Too Many Cats:
The Problem of the Many and the Metaphysics of
Vagueness
Nicholas K. Jones
Ph.D. Thesis
Birkbeck, University of London (School of Philosophy)
Submitted September 2010
1
Abstract
Unger’s Problem of the Many seems to show that the familiar macroscopic world is
much stranger than it appears. From plausible theses about the boundaries of or-
dinary objects, Unger drew the conclusion that wherever there seems to be just one
cat, cloud, table, human, or thinker, really there are many millions; and likewise
for any other familiar kind of individual. In Lewis’s hands, this puzzle was subtly
altered by an appeal to vagueness or indeterminacy about the the boundaries of
ordinary objects. This thesis examines the relation between these puzzles, and also
to the phenomenon of vagueness.
Chapter 1 begins by distinguishing Unger’s puzzle of too many candidates from
Lewis’s puzzle of borderline, or vague, candidates. We show that, contra Unger, the
question of whether this is a genuine, as opposed to merely apparent, distinction
cannot be settled without investigation into the nature of vagueness. Chapter 2 be-
gins this investigation by developing a broadly supervaluationist account of vague-
ness that is immune to the standard objections. This account is applied to Unger’s
and Lewis’s puzzles in chapters 3 and 4. Chapter 3 shows that, despite its popular-
ity, Lewis’s own approach to the puzzles is unsatisfactory: it does not so much solve
the puzzle, as prevent us from expressing them; it cannot be extended to objects
that self-refer; it is committed to objectionable theses about temporal and modal
metaphysics and semantics. Chapter 4 develops a conception of ordinary objects
that emphasises the role of identity conditions and change, and uses it to resolve
both Problems of the Many. This allows us to diagnose the source of the puzzles:
an overemphasis on mereology in contemporary material ontology.
2
I hereby declare that all the work presented in this thesis is my own:
Nicholas K. Jones
3
In loving memory of Richard Blundell
4
Acknowledgements
I have benefited greatly from the wisdom of my supervisors at Birkbeck: Dorothy
Edgington, Keith Hossack, Fraser MacBride and Ian Rumfitt all discussed drafts
(or ancestors of drafts) of various chapters. Without their feedback and encour-
agement this thesis would certainly have been much worse than it actually is. Of
everyone I’ve met since coming to Birkbeck, Dorothy and Fraser deserve special
thanks; their assistance at every stage of this project went far beyond that which I
had any right to expect.
Innumerable conversations with Will Bynoe have shaped many of the views
here; I’m grateful for his good sense and constant willingness to talk philosophy
(as well as film, science-fiction, boardgames and comics). The Birkbeck graduate
community provided a friendly environment in which to learn from my mistakes;
conversations with Alex Douglas, Simon Hewitt, Sam Lebens, Steven Methven,
Gina Tsang and Dan Turnbull stand out as particularly enjoyable and construc-
tive. Thanks to Mark Textor for feedback on part of chapter 2, and to the King’s
Metaphysics Society for providing an audience for early versions of many of the
arguments here. Thanks also to Rory Madden, Jonny McIntosh and Lee Walters.
I received funding from the Arts and Humanities Research Council during my
first three years at Birkbeck, and a Royal Institute of Philosophy Jacobsen Fellow-
ship whilst I was writing-up; thanks to both organisations.
I count myself lucky for the unceasing support of all my friends and family.
I’m especially grateful to Lara Freeman for competing with philosophy and always
having faith in me, as well as for putting up with a sometimes thoughtless and
distracted boyfriend. My greatest debt is to my parents, for everything they’ve
ever done for me.
5
Contents
Introduction 13
1 Two Problems of the Many 15
2 Supervaluationist Theories of Vagueness 64
3 Vagueness in Reference 137
4 Identity Conditions and Constitution 190
Bibliography 258
6
Analytical Contents
Introduction 13
1 Two Problems of the Many 15
1.1 Unger’s puzzle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.1.1 How many clouds? . . . . . . . . . . . . . . . . . . . . . . . . 16
1.1.2 Dispensing with fusions . . . . . . . . . . . . . . . . . . . . . 18
1.1.2.1 Fusion and lumps of matter . . . . . . . . . . . . . . 19
1.1.2.2 Fusions dispensed with . . . . . . . . . . . . . . . . 29
1.1.3 Principles of minute differences . . . . . . . . . . . . . . . . . 32
1.1.3.1 Two bad arguments against the principle of minute
differences . . . . . . . . . . . . . . . . . . . . . . . . 32
1.1.3.2 Two justifications for the principle of minute differ-
ences . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.1.4 Selection and exclusion principles . . . . . . . . . . . . . . . 35
1.1.4.1 The alternative argument . . . . . . . . . . . . . . . 35
1.1.4.2 Maximality . . . . . . . . . . . . . . . . . . . . . . . 36
1.1.4.3 Massive overlap . . . . . . . . . . . . . . . . . . . . . 39
1.1.4.4 Brutalism . . . . . . . . . . . . . . . . . . . . . . . . 39
1.1.5 Extension to other ordinary kinds . . . . . . . . . . . . . . . . 43
1.1.5.1 Ordinary objects and sorts . . . . . . . . . . . . . . . 43
1.1.5.2 The problem for non-clouds . . . . . . . . . . . . . . 44
1.2 Lewis’s puzzle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
1.2.1 Borderline parts . . . . . . . . . . . . . . . . . . . . . . . . . . 48
1.2.1.1 Vagueness . . . . . . . . . . . . . . . . . . . . . . . . 48
Contents 7
1.2.1.2 Mereological vagueness . . . . . . . . . . . . . . . . 52
1.2.2 Why many cats? . . . . . . . . . . . . . . . . . . . . . . . . . . 52
1.3 How many puzzles? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
1.4 Puzzle or problem? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
1.4.1 Time, modality and coincidence . . . . . . . . . . . . . . . . . 55
1.4.2 Causation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
1.4.3 Free will . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
1.4.4 Real choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
1.4.5 Moorean fact . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
1.4.6 Responsibility . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
1.4.7 Singular thought and reference . . . . . . . . . . . . . . . . . 62
1.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2 Supervaluationist Theories of Vagueness 64
2.1 Supervaluationist formal theory . . . . . . . . . . . . . . . . . . . . . 65
2.2 Understanding the supervaluationist formalism . . . . . . . . . . . . 67
2.2.1 Consequence, truth and interpretations . . . . . . . . . . . . 68
2.3 Interpretations as supervaluationist models . . . . . . . . . . . . . . 70
2.3.1 Particular . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.3.2 Subtruth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.3.3 Supertruth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.3.3.1 Sharpenings as ways a vague language could be made
precise . . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.3.3.2 Sharpenings as ways a precise boundary could be
drawn . . . . . . . . . . . . . . . . . . . . . . . . . . 76
2.3.3.3 Sharpenings as classical interpretations . . . . . . . 77
2.3.3.4 Sharpenings as theoretical posits . . . . . . . . . . . 78
2.3.3.5 Sharpenings as artefacts . . . . . . . . . . . . . . . . 79
2.3.4 Interpretations as models: concluding remarks . . . . . . . . 81
2.4 Interpretations as sharpenings . . . . . . . . . . . . . . . . . . . . . . 81
2.4.1 The association relation . . . . . . . . . . . . . . . . . . . . . 81
2.4.2 A Lewisian theory of association . . . . . . . . . . . . . . . . 82
Contents 8
2.4.3 Truth and consequence . . . . . . . . . . . . . . . . . . . . . . 85
2.4.4 An objection: monadic truth . . . . . . . . . . . . . . . . . . . 86
2.4.4.1 The monadicity of ‘is true’ . . . . . . . . . . . . . . . 86
2.4.4.2 The monadicity of Truth . . . . . . . . . . . . . . . . 87
2.4.5 The story so far . . . . . . . . . . . . . . . . . . . . . . . . . . 88
2.5 Four benefits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
2.5.1 Analysing clarity . . . . . . . . . . . . . . . . . . . . . . . . . 88
2.5.1.1 Clarity and the Supertruth View . . . . . . . . . . . 89
2.5.1.2 Clarity and the Sharpenings View . . . . . . . . . . 90
2.5.2 The Sorites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
2.5.3 Classical logic . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
2.5.4 Penumbral connection . . . . . . . . . . . . . . . . . . . . . . 94
2.6 A logical problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
2.6.1 The argument . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
2.6.2 Restricting CP . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
2.6.3 The justification for RCP . . . . . . . . . . . . . . . . . . . . . 99
2.6.4 A problem for RCP . . . . . . . . . . . . . . . . . . . . . . . . 101
2.6.5 The Sharpening View . . . . . . . . . . . . . . . . . . . . . . . 102
2.6.6 Supervaluationist logic: concluding remarks . . . . . . . . . 102
2.7 Two semantic problems . . . . . . . . . . . . . . . . . . . . . . . . . . 103
2.7.1 Truth and supertruth . . . . . . . . . . . . . . . . . . . . . . . 103
2.7.1.1 Supertruth and Bivalence . . . . . . . . . . . . . . . 103
2.7.1.2 From LEM to Bivalence . . . . . . . . . . . . . . . . 104
2.7.1.3 The Sharpening View . . . . . . . . . . . . . . . . . 109
2.7.2 Borderline discretion . . . . . . . . . . . . . . . . . . . . . . . 109
2.7.2.1 Supertrtuth and borderline discretion . . . . . . . . 109
2.7.2.2 Sharpenings and borderline discretion . . . . . . . . 110
2.7.3 Supervaluationist semantics: concluding remarks . . . . . . . 112
2.8 Field on truth and super-truth . . . . . . . . . . . . . . . . . . . . . . 112
2.9 Higher-order vagueness . . . . . . . . . . . . . . . . . . . . . . . . . . 114
2.9.1 Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
2.9.2 Varieties of higher-order vagueness . . . . . . . . . . . . . . . 116
Contents 9
2.9.3 Higher-order vagueness and the Supertruth View . . . . . . . 117
2.9.4 Semantics for higher-order vagueness . . . . . . . . . . . . . 118
2.9.5 Hidden sharp boundaries? . . . . . . . . . . . . . . . . . . . . 120
2.9.6 Sainsbury on vagueness and set-theoretic semantics . . . . . 123
2.9.7 Objection: the fragmentation of vagueness . . . . . . . . . . . 126
2.9.8 More hidden sharp boundaries? . . . . . . . . . . . . . . . . . 127
2.9.9 Higher-order vagueness and the Sharpening View . . . . . . 130
2.9.9.1 Metasemantic gradualness . . . . . . . . . . . . . . 130
2.9.9.2 Absolute clarity . . . . . . . . . . . . . . . . . . . . . 133
2.9.10 Higher-order vagueness: concluding remarks . . . . . . . . . 135
2.10 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
3 Vagueness in Reference 137
3.1 Two solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
3.1.1 Two options . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
3.1.2 Unger and Lewis . . . . . . . . . . . . . . . . . . . . . . . . . 141
3.1.3 One Cat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
3.1.4 Many Cats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
3.1.5 The Problem of the Two . . . . . . . . . . . . . . . . . . . . . 145
3.2 Four problems with vague reference . . . . . . . . . . . . . . . . . . 146
3.2.1 Schiffer on speech reports . . . . . . . . . . . . . . . . . . . . 146
3.2.1.1 Three problem cases . . . . . . . . . . . . . . . . . . 146
3.2.1.2 Diagnosis . . . . . . . . . . . . . . . . . . . . . . . . 148
3.2.1.3 Cure . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
3.2.1.4 Vague and precise contents . . . . . . . . . . . . . . 154
3.2.2 Barnett on incomplete definitions . . . . . . . . . . . . . . . . 158
3.2.2.1 Solution . . . . . . . . . . . . . . . . . . . . . . . . . 159
3.2.3 McGee and McLaughlin on de re belief . . . . . . . . . . . . . 161
3.2.3.1 The problem . . . . . . . . . . . . . . . . . . . . . . 161
3.2.3.2 Multiply interpreted de re belief . . . . . . . . . . . 163
3.2.4 Sorensen on Direct Reference . . . . . . . . . . . . . . . . . . 164
3.3 Three problems with the Problem of the Many . . . . . . . . . . . . . 166
Contents 10
3.3.1 A genuine solution? . . . . . . . . . . . . . . . . . . . . . . . . 166
3.3.1.1 McKinnon’s first horn: principled sharpenings . . . 167
3.3.1.2 McKinnon’s second horn: arbitrary sharpenings . . 168
3.3.1.3 Extensive overlap and intrinsicality . . . . . . . . . 168
3.3.1.4 Genuine solution or semantic trickery? . . . . . . . 170
3.3.2 Hawthorne on self-reference . . . . . . . . . . . . . . . . . . . 173
3.3.2.1 The argument . . . . . . . . . . . . . . . . . . . . . . 173
3.3.2.2 First response: deny (2) . . . . . . . . . . . . . . . . 177
3.3.2.3 Second response: deny (3) . . . . . . . . . . . . . . . 178
3.3.2.4 Third response: deny (4) . . . . . . . . . . . . . . . . 179
3.3.2.5 Fourth response: deny (5) . . . . . . . . . . . . . . . 179
3.3.3 Coincidence, time and modality . . . . . . . . . . . . . . . . . 180
3.3.3.1 Coincidence, persistence and modality . . . . . . . 181
3.3.3.2 Perdurance . . . . . . . . . . . . . . . . . . . . . . . 183
3.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
4 Identity Conditions and Constitution 190
4.1 Vagueness in parthood and constitution . . . . . . . . . . . . . . . . 191
4.1.1 Why only one cat? . . . . . . . . . . . . . . . . . . . . . . . . . 192
4.2 The Proposal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
4.2.1 Objects and change . . . . . . . . . . . . . . . . . . . . . . . . 195
4.2.2 Two kinds of identity criterion . . . . . . . . . . . . . . . . . . 198
4.2.3 The one-level proposal . . . . . . . . . . . . . . . . . . . . . . 199
4.2.3.1 Applying the one-level proposal . . . . . . . . . . . 201
4.2.4 The two-level proposal . . . . . . . . . . . . . . . . . . . . . . 203
4.2.4.1 Applying the two-level proposal . . . . . . . . . . . 208
4.2.5 How many levels? . . . . . . . . . . . . . . . . . . . . . . . . . 210
4.2.6 Matter and constitution . . . . . . . . . . . . . . . . . . . . . 211
4.2.7 Identity criteria: concluding remarks . . . . . . . . . . . . . . 211
4.3 Objections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
4.3.1 Unique Constitution . . . . . . . . . . . . . . . . . . . . . . . 213
4.3.1.1 First objection . . . . . . . . . . . . . . . . . . . . . . 213
Contents 11
4.3.1.2 Second objection . . . . . . . . . . . . . . . . . . . . 214
4.3.2 Cats and maximal lumps . . . . . . . . . . . . . . . . . . . . . 215
4.3.3 Fission and fusion . . . . . . . . . . . . . . . . . . . . . . . . . 215
4.3.4 Lewis on cats and cat-constituters . . . . . . . . . . . . . . . . 216
4.3.5 Stating the criteria . . . . . . . . . . . . . . . . . . . . . . . . 217
4.3.5.1 Stating a two-level criterion . . . . . . . . . . . . . . 217
4.3.5.2 Stating a one-level criterion . . . . . . . . . . . . . . 220
4.3.6 Identity and analysis . . . . . . . . . . . . . . . . . . . . . . . 221
4.4 Property-possession . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
4.4.1 Three kinds of property-inheritance . . . . . . . . . . . . . . 223
4.4.2 Against Universal Inheritance . . . . . . . . . . . . . . . . . . 224
4.4.3 Against Existential Inheritance . . . . . . . . . . . . . . . . . 225
4.4.4 In defence of Relativised Inheritance . . . . . . . . . . . . . . 225
4.4.4.1 A problem for RI . . . . . . . . . . . . . . . . . . . . 225
4.4.4.2 An instantiation relation? . . . . . . . . . . . . . . . 227
4.4.4.3 A relational account of inheritance . . . . . . . . . . 230
4.4.4.4 Non-relational Relativised Inheritance . . . . . . . 232
4.4.5 Property-possession: concluding remarks . . . . . . . . . . . 234
4.5 Vagueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
4.5.1 Unclarity in constitution, inheritance and parthood . . . . . 235
4.5.1.1 Unclarity in constitution . . . . . . . . . . . . . . . 235
4.5.1.2 Unclarity in inheritance . . . . . . . . . . . . . . . . 237
4.5.1.3 Unclarity in Parthood . . . . . . . . . . . . . . . . . 238
4.5.2 Constitutional vagueness? . . . . . . . . . . . . . . . . . . . . 239
4.5.2.1 Sorites-susceptibility and higher-order borderline cases240
4.5.2.2 A hidden assumption . . . . . . . . . . . . . . . . . 242
4.5.2.3 Three kinds of response . . . . . . . . . . . . . . . . 244
4.5.3 Content-determination without Eligibility . . . . . . . . . . . 244
4.5.3.1 Against Eligibility . . . . . . . . . . . . . . . . . . . 245
4.5.3.2 Constitutional vagueness without Eligibility . . . . 246
4.5.4 Limiting constitutional unclarity . . . . . . . . . . . . . . . . 247
4.5.4.1 First challenge: the Sorites . . . . . . . . . . . . . . 248
Contents 12
4.5.4.2 Second challenge: borderline precision . . . . . . . 250
4.5.4.3 Third challenge: generalisation to other cases . . . . 250
4.5.5 Gradual constitution . . . . . . . . . . . . . . . . . . . . . . . 251
4.5.5.1 The proposal . . . . . . . . . . . . . . . . . . . . . . 251
4.5.5.2 A limit on higher-order borderline cases? . . . . . . 253
4.5.5.3 Relativised property-possession . . . . . . . . . . . 254
4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
Bibliography 258
13
Introduction
Unger’s Problem of the Many seems to show that almost all of our ordinary numer-
ical judgements are radically mistaken: for any kind K of ordinary macroscopic
object, wherever there appears to be just one K, really there are many millions.
Lewis presented a similar argument, though he appealed to vagueness or indeter-
minacy, whilst Unger did not. This thesis investigates the relation between these
puzzles, and their connection to the phenomenon of vagueness.
Chapter 1 develops versions of the puzzles that require only very minimal as-
sumptions. We show the puzzles do not primarily concern the existence of individ-
uals, but the instantiation of ordinary sortal properties. Puzzles about the number
of objects arise only via the commonplace assumption that each instantiation of an
ordinary sortal is by a single individual; even those who deny that ordinary com-
posite individuals exist must address the Problem of the Many. We will also see
that, contra Unger, the question of whether he and Lewis were addressing the same
puzzle cannot be settled prior to an investigation into the nature of vagueness. The
chapter closes by arguing that there are good reasons to reject Unger’s conclusion:
our ordinary numerical judgements are not typically radically mistaken. The con-
clusion of Unger’s puzzle is not merely implausible—surely there are not many
humans seated in my chair, writing this thesis—but creates significant problems
in the metaphysics of time, modality, free will, choice, moral responsibility and
singular thought.
Chapter 2 turns to vagueness, and supervaluationist theories of vagueness in
particular. We develop two broadly supervaluationist accounts of vagueness. One
treats vagueness as a semantic phenomenon, and uses classes of sharpenings to rep-
resent vague semantic structures. The other treats vagueness as a partly semantic
Introduction 14
and partly metasemantic phenomenon; classes of sharpenings are used to represent
classes of semantic structures that fit the meaning-determining facts well enough
to count as the actual, or intended, semantic structures of a vague language. On
this second view, vague languages express many precise contents. We argue that
the standard objections to supervaluationism do not touch the second view, and
hence that it is preferable to the first. That view is applied to the Problem of the
Many in chapters 3 and 4. Due to restrictions on space, a study of a recent argu-
ment for the classicality of supervaluationist logic, due to J.R.G. Williams, has been
removed, and is now forthcoming in the Journal of Philosophy.
Chapter 3 serves two purposes. The first is an examination of Lewis’s solution
to his puzzle. Lewis claimed that his and Unger’s puzzles are sources of referential
vagueness in our names for ordinary objects, and used the supervaluationist tech-
nique to ensure that sentences that express our ordinary numerical judgements
are true. We show that this approach: does not solve the problems, but merely
prevents us from expressing them; cannot be extended to self-referrers; is commit-
ted to objectionable theses about the metaphysics and semantics of temporal and
modal discourse. The second goal of the chapter is to defend the supervaluation-
ist view developed in chapter 2 against objections due to Schiffer, Barnett, McGee
and McLaughlin, and Sorensen. These objections all concern supervaluationist ac-
counts of vague reference.
Chapter 4 closes by developing a response to Unger’s puzzle and then extending
it to Lewis’s. We begin with the following thesis about ordinary objects: change
and persistence are explanatorily prior to mereology, constitution and location. We
use this thesis to argue that Unger’s puzzle shows only that ordinary macroscopic
objects may not have a unique collection of microscopic constituents, not that there
are many such objects where there appears to be only one: a single ordinary object
may be simultaneously constituted by several (partially disjoint) portions of matter.
Having developed two versions of this view, we close with three ways of extending
it to Lewis’s puzzle of constitutional vagueness.
15
Chapter 1
Two Problems of the Many
In “The problem of the many”, Peter Unger presented a novel and intriguing puzzle
about ordinary macroscopic material objects (Unger, 1980). From plausible theses
about the boundaries of these objects, Unger drew the conclusion that our ordinary
numerical judgements about them are radically mistaken. David Lewis addressed
this in his “Many, but almost one” (1993a). But in Lewis’s hands the problem was
subtly altered by an appeal to vagueness or indeterminacy in the boundaries of
ordinary objects. It is therefore not clear whether Unger and Lewis were addressing
the same problem. The issue turns in part upon the nature of vagueness. One
goal of this thesis is clarity about the relation between these puzzles, and also the
phenomenon of vagueness. Another is greater clarity about the metaphysical and
semantic commitments of potential solutions to the puzzle. Yet another is to try
and solve the puzzles. The first step towards achieving these goals is to find out
what the puzzles are. That is the purpose of this chapter.
Unger’s puzzle is presented in §1.1, and Lewis’s in §1.2. We turn to their rela-
tionship in §1.3. §1.4 closes by asking whether these are genuine problems or mere
puzzles. The next chapter develops a supervaluationist approach to vagueness.
Different applications of this approach to the Problem of the Many are examined
in chapters 3 and 4.1
1 Peter Geach presented a similar puzzle in §110 of the third edition of Reference and Generality,
which he attributed to William of Sherwood (Geach, 1980). A recent ancestor of this thesis contained
an examination of the relation between Geach’s puzzle and those of Unger and Lewis, as well as
Geach’s doctrine of Relative Identity. The result of including that discussion was however, far too
Two Problems 16
1.1 Unger’s puzzle
This section presents Unger’s puzzle. Although versions arise for all ordinary kinds
of macroscopic object, our initial presentation and discussion follows Unger in fo-
cusing upon clouds. It is comparatively easy to see how Unger’s puzzle arises for
clouds, despite it being perhaps somewhat doubtful whether they are a genuine
kind of individual.
The initial presentation of the puzzle is in §1.1.1. §1.1.2 clarifies the puzzle and
weakens its ontological assumptions. A key premiss is examined in §1.1.3, and an
alternative provided in §1.1.4. §1.1.5 closes by extending the puzzle from clouds
to other ordinary sorts of object.
1.1.1 How many clouds?
Unger begins by asking us to consider a typical cloud, C. Let a be a water molecule
on C’s left-hand boundary;2 let b be a water molecule in C’s exterior but extremely
close to its right-hand boundary. Consider the object D that differs from C only by
excluding a and including b:
Cam bm Dam bmUnger asks: is D a cloud? He answers: yes.
Not just anything is a cloud; an appropriate internal structure and constitution
are required. Let resemblance in cloud-respects be resemblance w.r.t. this structure
and constitution. Unger (1980, p.122) endorses the following principle of minute
differences:
PMD1 If x is a typical cloud and y differs only minutely in cloud-respects from x,
then y is a cloud.3
long a thesis; so it was removed shortly prior to submission.2 We assume that cloud interiors are closed: droplets on a cloud’s boundary are in its interior.
Nothing turns on this.3 Initial universal quantifiers will often be omitted in the interests of readability.
Two Problems 17
Since C is a typical cloud and D differs only minutely in cloud-respects from C,
it follows that D is a cloud. Since C and D each has a part, the droplets a and b
respectively, that the other does not, they are distinct clouds.
Why grant that D differs only minutely in cloud-respects from C? Cloud-
interiors are characterised by a high density of suspended water droplets, and
cloud-exteriors by a low density thereof.4 The transition from high density interior
to low density exterior is gradual, not marked by any sharp fall in droplet-density.
This ensures that we can select droplet b from sufficiently close to C’s boundary to
guarantee that D differs only minutely in cloud-respects from C. Since the bound-
aries of all typical clouds are like this, we can generalise: for any typical cloud,
another cloud is almost co-located with it.
We need not stop at two clouds. Millions of droplets lie on the boundary of
each typical cloud. And millions of droplets in each typical cloud’s exterior are
extremely close to its boundary. Any two such would suffice in place of a and b.
There is one cloud for each of these pairs. So at least 1012 clouds are almost co-
located with each typical cloud. We could even have included or excluded two
(or three, or. . . ) droplets and still obtained objects that differ minutely in cloud-
respects from C. So really there will be even more clouds than this.5
How can this be? The intuitive view is that, sometimes, a typical cloud is the
only cloud in the sky. But from this premiss, Unger’s argument leads to the con-
tradictory conclusion that, in those situations, there are billions of clouds in the
sky. At best, our numerical judgements about clouds are in radical error: there
are vastly many more than we thought. Unger thinks that if clouds exist, then our
ordinary numerical judgements about them are typically correct. He sees this as
a non-negotiable component of our ordinary world-view.6 So, he claims, clouds
4 We assume that only water droplets are parts of clouds. Nothing turns on this simplification.5 Droplets slightly further into C’s interior and exterior would presumably have been acceptable
too.6 Unger’s most recent work on the puzzle rejects this (Unger, 2006a, ch.7). It is unproblematic,
he thinks, for millions of clouds, tables, plants, and maybe even cats, to almost coincide where there
seems to be only one. But he does regard it as non-negotiable that he is the only conscious being in his
immediate vicinity, and he thinks that each of us will think the same about ourselves. He concludes
that only a form of mind-body substance dualism can respect this.
Two Problems 18
cannot exist: nothing can satisfy our concept cloud.
1.1.2 Dispensing with fusions
The section examines and weakens the ontological assumptions of Unger’s argu-
ment. Our original presentation assumed that some object D differs from C only
by (i) not including one of C’s boundary-droplets, and (ii) including one extremely
close droplet in C’s exterior. Why think that there is such an object? Does Unger’s
argument require it?
D’s existence is plausible, if D is conceived as a lump or portion of matter.
§1.1.2.1 examines this suggestion. We’ll see that this creates trouble for Unger’s
use of PMD1: it’s doubtful whether any mere portion of matter closely resembles
any cat in cat-respects. Three responses to this objection will then be considered: a
deviant temporal and modal semantics; a very liberal theory of matter; a modified
principle of minute differences. The third is most satisfactory. But even this is
committed to the existence of arbitrary lumps of matter, and that assumption is not
beyond reproach. §1.1.2.2 invokes the apparatus of plural logic to show that D’s
existence is an inessential assumption. We thereby strengthen Unger’s argument
by maximally weakening its ontological assumptions.
Some terminology will aid our discussion. Using the notion of improper
parthood—the sense of parthood in which everything is a part of itself—we define:
x overlaps y iff something is part of both x and y.
x is disjoint from y (also: x excludes y) iff x does not overlap y.
x includes y iff everything that overlaps y also overlaps x.
x is a fusion of set s (also: x fuses s; s composes x) iff (i) everything that overlaps
x overlaps some member of s, and (ii) everything that overlaps some member
of s overlaps x.
Let s0 be the set of water droplets that composes our typical cloud C; let sB be the
set of droplets on C’s boundary; let sE be the set of droplets only just in C’s exterior;
let s1, . . . , sn be all the sets whose members are (i) every member of s0 except some
one member of sB, and (ii) some member of sE; let each Di amongst D0, . . . , Dn be a
Two Problems 19
fusion of si. Since there are millions of droplets in sB and sE, there are millions of
cloud-candidates Di.
The argument for millions of clouds assumes that each si has a fusion Di. The
principle of minute differences PMD1 is then invoked to conclude that each Di is a
cloud. Why grant that such fusions exist? Does Unger require them?
1.1.2.1 Fusion and lumps of matter
Does every candidate Di exist? The following entails that they do:
Unrestricted Fusion Every set has a fusion.
Although Unrestricted Fusion has its defenders, notably Lewis (1991) and Theodore
Sider (2001a), it is highly controversial. Despite this controversy, there are entities
for which it is plausible: lumps (or portions) of matter, masses and space-time
points.7 This section considers the following question: can we take the Di’s in
Unger’s argument as lumps of matter? If so, then the result is a version of the
argument in which the existence of the candidates isn’t overly controversial.
We begin with a problem for this account of the candidates: since no lump
and cloud resemble one another closely in cloud-respects, PMD1 doesn’t imply
that any of the candidates is a cloud. Three kinds of response will be examined.
The first invokes a non-standard temporal and modal semantics in order to block
the argument against clouds and lumps resembling one another in cloud-resects.
The second response invokes a more liberal theory of matter. Both these responses
will be rejected. The section closes with a more satisfying response that invokes a
different principle of minute differences. The next section shows how to do without
the existence of individual candidates entirely.
Matter and minute differences Material objects are, in some sense, made out
of matter. Unrestricted Fusion is an attractive thesis about matter. To see this,
consider any collection of material objects. Some (lump of) matter makes them
7 Throughout, we use ‘lump’ in a semi-technical sense for a porion of the material “stuff” from
which ordinary macroscopic objects are made, whatever that stuff might ultimately turn out to be.
Two Problems 20
up.8 That lump is a fusion of the set of those objects. So Unrestricted Fusion holds
for lumps.9 Taking each candidate Di as the lump from which the members of
the set si are made (and hence as a fusion of that set), allows us to appeal to this
intuitively attractive argument in support of their existence.
This approach is problematic. Following are two reasons to doubt that any
cloud and lump of matter resemble one another at all closely in cloud-respects.
Each is a reason to doubt therefore, that the principle of minute differences PMD1
implies that any lump Di is a cloud.
First reason: lumps have only permanent and necessary parts, while clouds do
not.
x is a permanent part of y iff x is part of y at every time at which y exists.
x is a necessary part of y iff, necessarily, if y exists, then x is part y.10
Lump-mereology is modally and temporally invariant. Since clouds can have dif-
ferent droplet-parts at different times and could have had different droplet-parts
than they actually do, clouds differ significantly from lumps in cloud-respects.
Second reason: clouds and lumps have different existence and identity condi-
tions. For example, a lump exists iff its constituent sub-portions of matter exist,
but clouds exist only when their droplets are sufficiently densely arranged. Since
no lump and cloud have the same existence conditions, no lump closely resembles
a cloud in cloud-respects.
A successful reduction of ordinary objects to lumps of matter would avoid both
these problems. There are two strategies this reduction might take. The first in-
8 Note the collective reading here: for any objects, some matter makes them up without making
any one of them up (unless “they” are one).9 The assumption that, for any objects, some lump makes them up is essential here, and tanta-
mount to our conclusion. The point is not to provide independent argument for Unrestricted Fusion,
but merely to illustrate how natural it is for lumps.10 Interestingly, masses seem to be unlike lumps of matter in this respect: only some parts of
masses need be permanent or necessary, specifically, those of the kind of which it is a mass. If x is a
mass of water, then x’s water molecule parts are permanent and necessary parts of x. But the parts
of those water molecules may be neither permanent nor necessary parts of x, despite being parts of
x. For water molecules can survive changes in their constituent electrons without endangering the
existence of any water-mass of which they are parts. See Zimmerman (1995) for more.
Two Problems 21
vokes a non-standard temporal and modal semantics. The second invokes a more
flexible theory of matter. We consider and reject these strategies in turn. We
then present an alternative non-reductive strategy involving a different principle
of minute differences.
First strategy: counterpart-theory The problem with taking the candidates Di
as lumps of mater was that clouds and lumps have different modal and temporal
profiles. We now examine the use of counterpart-theory to block the argument
from modal and temporal differences to non-identity. If successful, this will allow
us to maintain that clouds and lumps satisfy exactly the same modal and temporal
formulae, and hence undermine the argument for significant differences in cloud-
respects between clouds and lumps.
David Lewis introduced counterpart-theory as an account of de re modal predi-
cation (Lewis, 1968, 1971, 1986b). Although Lewis formulated it as an extensional
translation of de re modal discourse, not as a semantic theory for a modal language,
it is reasonably clear how to obtain a semantic theory from it (Hazen, 1979; Stal-
naker, 1986, 1994, dicuss some of the issues). The result departs from standard
possible-worlds style modal semantics in four ways:
(i) Distinct worlds have disjoint domains: nothing exists in more than one world.11
(ii) There is a collection of binary counterpart relations R that hold only between
individuals in distinct worlds.
(iii) Counterpart relations are similarity relations.
(iv) The satisfaction of modal formulae by objects is (a) relativised to counterpart
relations and (b) determined by the satisfaction of the corresponding non-
modal formulae by their counterparts:
x satisfies p�Aq relative to counterpart relation R iff everything x bears
R to (its R-counterparts) satisfies A.
11 At least, no particulars wholly exist in more than one world: multiply located universals, if there
are such things, would be wholly present in multiple worlds.
Two Problems 22
Closed sentences are evaluated for truth by selection of an appropriate counterpart
relation.
We want to interpret modal talk counterpart-theoretically to make sentences
like ‘something is not a necessary part of Tim’ true, despite ‘Tim’ referring to a
lump of matter. This allows us to drop theses (i) and (iii) of Lewisian counterpart-
theory (though we are not compelled to), modifying (ii) and (iv) thus:
(ii′) There is a collection of four-place counterpart relations R: x in world w, is an
R-counterpart of y in world w′.12
(iv′) x satisfies p�Aq at world w relative to counterpart relation R iff, for any object
y and world w′, if y in w′ is an R-counterpart of x in w, then y satisfies A in
w′.13
On this view it can be true that some of Tim’s parts are not necessary parts, despite
‘Tim’ referring to a lump l of matter. The reason is that l’s Tim-counterparts not
have the same parts as it(s lump-counterparts): that x in w is an R-counterpart of
y in w′, does not imply that x = y.
Replacing worlds with times gives a temporal version of counterpart-theory. On
this view, it can be true that Tim has some non-permanent parts, despite ‘Tim’ re-
ferring to a lump l of matter. The reason is that l’s past and future Tim-counterparts
need not have the same parts as it(s lump-counterparts): that x at time t, is a tem-
poral R-counterpart of y at t′, does not imply that x = y.
Say that an object persists iff it exists at more than one time. One form of
temporal-parts theory, namely perdurance-theory, identifies ordinary persistents
with fusions of momentary objects. Another version, stage-theory, identifies ordi-
nary persistents with momentary objects themselves (Sider, 2001a; Hawley, 2001).
In order to make ordinary de re temporal discourse true, stage-theory needs (some
variant on) temporal counterpart-theory. It should be noted however that tempo-
ral counterpart-theory does not require temporal-parts theory. Without temporal-
parts, the view is akin to Roderick Chisholm’s (1976, ch.3) theory of entia successiva:
12 Counterpart-relations are four-place because (a) lumps exist in more than one world, and (b) a
single lump might be Tim in one world, Tom in another, and Tim and Tom’s modal properties differ.13 This needs complicating to permit non-trivial iterated modality, but it will do as a first pass.
Two Problems 23
the relation by which we trace the path of an ordinary object through time is not
identity, but a continuity relation connecting different (mereologically invariant)
objects at different times. (Chisholm also endorses the additional thesis that ordi-
nary mereologically variable persistents are logical fictions.)
Given counterpart-theory, differences in the truth-values of modal and tempo-
ral claims about lumps and clouds can be attributed to differences in the coun-
terpart relations relative to which those claims are evaluated, rather than the sub-
jects of those claims. To the counterpart-theorist, differences in the truth-values of
modal claims about clouds and lumps therefore don’t show lumps are not clouds,
or that lumps and clouds don’t closely resemble one another in cloud-respects. But
why should we be counterpart-theorists? What are the motivations for counterpart-
theory? Such radical departure from standard modal and temporal semantics is
ill-motivated, if its only purpose is to allow retention of a matter-only ontology.
Sider (2001a, ch.5) argues that stage-theory, and hence counterpart-theory, pro-
vides the best unified response to the so-called “paradoxes of coincidence”: appar-
ent cases in which several objects fill and fit within the same region at the same time
(or even throughout time). But it is far from clear why coincidence is supposed to
be problematic. Lewis (1986b) has a different motivation for modal counterpart-
theory. He is forced into it by his ontology of concrete possible worlds. He also
argues that it is a component of the most satisfactory solution to a wide range of
puzzles. But Lewis’s ontology of concrete possible worlds controversial, to say the
least. And there are alternative responses to all of the puzzles Lewis addresses. A
counterpart-theoretic defence of Unger’s puzzle will be of very limited interest, if
these are the motivations for counterpart-theory. I do not claim that these are the
only motivations for counterpart-theory. They are however, some of the best and
most prominent.
Setting worries about its motivation aside, all forms of counterpart-theory face
similar objections. Firstly, counterpart-theory implies that our ordinary judge-
ments of cross-time or cross-possibility sameness are not judgements of identity.
Secondly, and relatedly, utterances of de re predications at different times have dif-
ferent subjects: the present truth-condition of ‘Nick is typing’ is that a certain ob-
ject is typing, while five minutes hence it will be that some other object is typing.
Two Problems 24
Thirdly, and most significantly, it is doubtful whether modal counterpart-theory is
consistent (Stalnaker, 1986, §2). The present version denies that there are possibili-
ties where you have any parts other than your actual parts, whilst also maintaining
that you could have had different parts than you actually do. Without an account of
possibilities other than as possible ways things could be, or similar, this is contradic-
tory. The problem concerns the appearance of modal vocabulary in the analysis of
possibilities. Counterpart-theory thus brings commitment to a non-modal reduc-
tion of modality. It is rightly controversial whether any such reduction is possible.
Temporal counterpart-theory does not obviously suffer this last objection because
our access to non-present times may not be mediated by temporal vocabulary in
quite the same way as our access to non-actual possibilities is mediated by modal
and counterfactual vocabulary: we can remember the past, but not mere possibili-
ties. (For scepticism about this difference, see Edgington, 2010, §5).14
In light of these difficulties and its controversial motivations, let us set
counterpart-theory aside.
Second strategy: liberalism about lumps The problem with taking the candi-
dates Di in Unger’s argument as lumps of matter is that lumps and clouds have
different temporal and modal properties. This section presents and rejects a more
liberal conception of matter on which this is not the case.
Consider the following (partial) theory of matter:
For every filled spatiotemporal region r, some lump occupies exactly the
points in r.
For each cloud, this principle delivers a lump of matter that occupies exactly the
same points as it throughout the cloud’s history. Since this lump and cloud have
the same temporal profile, that provides no bar to identifying them. The objection
to the use of PMD1 in Unger’s argument therefore fails.
14 Even if counterpart-theory is defensible, Fine (2003) highlights significant non-modal and non-
temporal differences between ordinary objects and lumps: a statue, unlike its constituent lump, may
be Romanesque; a cat, unlike its constituent lump, may purr. Counterpart-theory does not address
these cases.
Two Problems 25
This is not our ordinary conception of matter. Suppose that a certain statue has
exactly the same parts throughout its existence. Does a lump come into existence
when the statue does, occupy the same space as the statue throughout the first half
of the statue’s history and then cease to exist? Insofar as our ordinary conception of
matter speaks to this question, the answer seems to be negative. Yet the principle
above entails a positive answer. That principle does not govern our intuitive con-
ception of matter and can therefore provide no intuitive support for the existence
of the candidates in Unger’s argument. Defending Unger’s puzzle by appeal to
controversial theses about matter limits its interest; it becomes a puzzle for certain
theories of objects, not for our ordinary world-view.
There is a second problem for lumps governed by the principle above: it is
silent about their modal profiles. If lumps have only necessary parts, then lumps
and clouds will still differ significantly in modal respects relevant to their being
clouds. But if we allow lumps to have some merely contingent parts, then the sense
in which we have really reduced objects to lumps of matter is unclear; this looks
more like a reduction of matter to objects (in combination with a plenitudinous
view of objects). The plausibility of Unrestricted Fusion for matter can then lend
no intuitive support to the existence of the candidate Di’s.
A response to both these worries is available. Our ordinary conception of matter
is inegalitarian: every way of carving up a filled region into subregions is equally
legitimate, in the sense that, for each such sub-region r, some lump fills and fits
within r. We can respond to the worries above by taking this inegalitarianism as an
analysis of our ordinary conception of matter and combining it with a liberal view
of regions, to give:
For every function f from worlds w onto filled spatiotemporal regions of w,
there is a lump that, in w, occupies all and only the points in f (w).15,16
This provides provides, for each cloud c, a lump of matter that is necessarily co-
located with c. Differences in modal and temporal properties provide no bar to
15 Treat regions as sets of points. When f (w) = ∅, f defines an object without spatiotemporal lo-
cation in w. Whether this object exists in w depends on whether spatiotemporal location is necessary
for being something.16 The Plenitude Lover in Hawthorne (2006c) endorses this thesis.
Two Problems 26
the identification of such lumps with their coincident clouds. The objection to
Unger’s use of PMD1 then fails. However, this response still rests on a controversial
analysis of matter. A version of Unger’s puzzle is of limited interest if it relies on
this analysis to defend the existence of the candidates Di.
A slightly different worry afflicts both strategies. Matter is extensional: whether
there is at least one lump for every filled spatial region, or filled spatiotemporal re-
gion, or function from worlds onto filled spatiotemporal regions, there is no more
than one. Consider the original, and simplest, account of lumps, and the view
that identifies ordinary objects with them. Let n be the lump that is now Nick.
Then the object-language argument from ‘Nick used to be made of different mat-
ter’ and ‘it is not the case that n used to be made of different matter’ to ‘Nick 6= n’
must be invalid. Kit Fine (2003) argues that this has untenable consequences in the
philosophy of language. And Fine (2000) presents an example of two necessarily
co-located objects of which different things are (apparently) true. On each theory
of lumps canvassed here, the corresponding argument from these predicative dif-
ferences to the distinctness of these objects must be invalid. Each of these theories
of lumps therefore incurs these untenable consequences, if Fine’s arguments are
sound. This provides reason to be sceptical of any reduction of objects to mat-
ter, and hence also sceptical of any defence of Unger’s use of PMD1 that appeals
to such a reduction to defend the candidates Di and cloud C are closely alike in
cloud-respects. In the absence of a detailed examination of Fine’s arguments, this
is certainly not conclusive. We won’t however undertake that examination here
because an alternative defence of Unger’s argument is available.
Third strategy: modify the principle of minute differences We’ve rejected non-
standard semantic theories and analyses of matter as ways of defending Unger’s
argument against the claim that no cloud and lump are closely alike in cloud-
respects. This section presents a version of that argument that accepts these differ-
ences and employs an alternative principle of minute differences instead.
First we need a new dyadic relation: constitution. This is the relation between
a lump and whatever it makes up, or constitutes, e.g.: between the matter of your
body and your body, or the marble from which a statue was carved and the statue.
Two Problems 27
Begin by defining two notions of coincidence:
x materially coincides with y iff x and y both fuse some set s.
x spatially coincides with y iff x and y both fill and fit within the same region
of space.
The following biconditional may well be extensionally correct, whichever form of
coincidence it employs:
x constitutes y iff x coincides with y and x is a lump of matter.17
But intuitively, objects coincide because they are made out of the same matter. So
we should resist taking the right to analyse the left.
Recall our earlier introduction of resemblance in cloud-respects as resemblance
w.r.t. having a structure and make-up appropriate to clouds. Similarly, let resem-
blance in cloud-constituting respects be resemblance w.r.t. having a structure and
make-up appropriate to constituting a cloud. Here is a second principle of minute
differences:
PMD2 If (i) some typical cloud x is a fusion of a set s, (ii) x is constituted by y, and
(iii) some fusion f of a set s′ differs only minutely from y in cloud-constituting
respects, then: some fusion of s′ is a typical cloud (constituted by f ).
Now, (i) C is a fusion of s0, (ii) C is constituted by the lump D0, which is a fusion of
s0, and (iii) some fusion of each si, namely the lump Di, differs only minutely from
D0 in cloud-constituting respects. So by PMD2: some fusion of each si is a cloud
(and constituted by lump Di). The objection to the argument from PMD1 fails
because we’re now comparing lumps with lump in respects relevant to their con-
stituting clouds, rather than comparing lumps with clouds in respects relevant to
their being clouds. Granting that lumps and clouds are not alike in cloud-respects,
PMD2 implies only that each set si has some fusion that is a cloud, not that this
cloud is the lump Di.
To avoid trivialising this principle, and thereby rendering it dialectically inef-
fective, resemblance in cloud-constituting respects must be restricted to exclude,
17 How could this be false? A lump would have to coincide with an object it didn’t constitute (or
that wasn’t made from it). This does not appear to be a genuine possibility.
Two Problems 28
for example, resemblance w.r.t. constituting a cloud. It is hard to state the restriction
precisely. A restriction to microphysical properties and relations might suffice. But
what exactly are microphysical properties?18 Still, it seems that some such restric-
tion is possible. So let us henceforth simply assume that this is so. (§1.1.4 presents a
version of the puzzle that does not require this assumption; see especially §1.1.4.4.)
E.J. Lowe and Mark Johnston both appeal to differences in category between
lumps and objects in response to Unger’s puzzle: no fusion Di is a cloud because
each is a mere lump and no cloud is a lump (Lowe 1982a,c,b, 1995; Johnston 1992).
But which lump constitutes C? Lowe and Johnston reply that constitution is vague
and flesh this out along broadly supervaluationist lines. We return to supervalua-
tions in chapter 2 and constitutional vagueness in chapter 4 (and Lowe and John-
ston’s proposal in §4.1). In the meantime, it suffices to note two reasons why the
cloud/lump distinction alone cannot solve the problem. (i) It provides no reason
to doubt PMD2, and hence no reason to doubt that each Di constitutes a cloud. (ii)
It does nothing to show how, given that each Di does constitute a cloud, they could
all constitute the same cloud.
We now have in place a version of Unger’s argument that takes the candidate
Di’s as lumps of matter, and uses the principle PMD2 to conclude that each of
them constitutes a cloud. This version avoids our objection to the argument that
uses PMD1 because it does not assume that any cloud and lump are alike in cloud-
respects, or that any lump is a cloud. It also avoids appeal to non-standard se-
mantic theories or analyses of matter. This argument does however, assume that
each candidate Di exists; it doesn’t, then PMD2 doesn’t imply that it constitutes a
cloud. That these candidates do exist follows from Unrestricted Fusion for matter,
but although that principle is plausible, it is certainly not beyond reproach. The
next section presents a version of Unger’s argument that dispenses with the indi-
vidual Di’s altogether. This shows that Unger’s puzzle is not a puzzle about fusion,
18 Merricks (2003, ch.2, §§II–IV) takes an argument akin to the Problem of the Many to refute
the supervenience of mental properties (and composition) on the intrinsic microphysical properties
and relations of collections of atoms. We make no assumptions about intrinsicality. So Merricks’s
argument is silent about PMD2.
Two Problems 29
constitution or the existence of individuals, but about the instantiation of ordinary
sortal properties.
1.1.2.2 Fusions dispensed with
This section presents a version of Unger’s argument that does not assume the ex-
istence of any controversial entities. (We won’t even assume that any of the sets si
has a fusion.) So, what might we replace lumps of matter with, whose existence is
uncontroversial? The most obvious candidates are sets (though they’re existence
isn’t quite uncontroversial). We will need a new principle of minute differences:
PMD3 If some fusion of a set s is a typical cloud and some set s′ differs only
minutely from s in cloud-respects, then some fusion of s′ is a cloud.
We need to understand resemblance in cloud-respects here as resemblance amongst
sets w.r.t. those properties of sets that (non-trivially) determine whether they are
fused by a cloud. Orthodoxy implies that these will be extrinsic properties because
orthodoxy takes sets to lack spatiotemporal location.19 Intrinsic change requires
spatiotemporal location. So if resemblance in cloud-respects between sets were in-
trinsic, then if some cloud fused a set s at some time, then, at every time, some
cloud would fuse s.20 This is clearly not so.
Resemblance amongst sets in cloud-respects is extrinsic, and therefore does not
concern how those sets are in themselves. It obtains because of some other resem-
blances that obtain between some other entities. Which resemblances, and which
entities? The natural answer is:
s resembles s′ in cloud-respects to the degree that the members of s resemble
those of s′ w.r.t. their making up a cloud.
Two questions arise. Firstly, what is this talk of “the members of s”? And secondly,
what is it for those members to make up a cloud?
19 A property if extrinsic iff it is not intrinsic. A property is intrinsic iff it concerns how an object
really is, “considered in itself”. The intended contrast is between the extrinsic being an uncle and the
intrinsic being a man. The proper analysis of intrinsicality is a vexed issue we will not enter into here.20 The problem is even worse if sets are necessary existents.
Two Problems 30
The obvious answer to the first question is that this is plural talk: ‘the members
of s’ denotes (plurally) every member of s and nothing else. Following George Boo-
los (1984), let us take plural expressions to denote several members of the domain
from which the denotations of singular expressions are drawn. Plural expressions
do not denote plural individuals, but plurally denote many individuals.21
With this in place, we can answer our second question above by extending fu-
sion to plurals:
x is a fusion of the y’s (also: x fuses the y’s; the y’s compose x) iff (i) every-
thing that overlaps x overlaps (at least) one of the y’s, and (ii) everything that
overlaps any of the y’s overlaps x.
Some things make up a cloud iff it is a fusion of them. Note also that the second
argument-place of plural fusion is collective:
F is distributive iff, necessarily, if the y’s are F, then each y is F.
F is collective iff F is not distributive.
Other collective properties include being arranged in a circle and carrying the boat.
Given a resemblance relation amongst pluralities that is collective in both argu-
ment places, our initial gloss on resemblance amongst sets in cloud-respects be-
comes:
s resembles s′ in cloud-respects (to degree d) iff the members of s resemble the
members of s′ in respects relevant to their composing clouds (to degree d).
This relation is extrinsic because it holds amongst sets in virtue of the (intrinsic)
properties and relations of their members, not those of the sets themselves (and
sets can be intrinsically invariant despite intrinsic variation in their members).
The appeal to sets here is clearly redundant. Let resemblance in cloud-respects
amongst collections be resemblance amongst collections in respects relevant to
their composing clouds. We can now state a fourth principle of minute differences:
21 Subsequent talk of collections and pluralities should be understood as grammatically singular
but semantically plural talk about the objects amongst those pluralities and collections.
Two Problems 31
PMD4 If the x’s compose a typical cloud and the y’s differ only minutely in cloud-
respects from the x’s, then the y’s also compose a cloud.
This is no less plausible than any of PMD1–3. Let the si’s be the members of set si.
For each i, the si’s resemble the s0’s extremely closely in cloud-respects. Since the
s0’s compose our typical cloud C, PMD4 implies that the si’s compose a cloud Di.
When i 6= j: Di 6= Dj because each overlaps a water droplet disjoint from any the
other overlaps. So there are millions of clouds where we thought there to be just
one, each nearly coincident with our original cloud C.
The dispensability of assumptions about fusion shows that the Problem of the
Many cannot be solved simply by denying the existence of various individuals or
restricting fusion. In fact, we can show that it is not even primarily a problem
about the existence of ordinary individuals at all, but about the ordinary kinds or
sorts to which those objects belong. Suppose the following is true:
Compositional Nihilism Only singletons have fusions (and then only in the trivial
sense that every object x overlaps exactly those things that overlap x, includ-
ing x itself).
It does not follow that clouds do not exist, only that if they do, then either (i) they
have no proper-parts, or (ii) cloud is collectively instantiated by the pluralities of
objects that we would ordinarily say compose clouds. Consider (ii) and the follow-
ing principle of minute differences:
PMD5 If the x’s are collectively a typical cloud and the y’s differ only minutely in
cloud-respects from the x’s, then the y’s are also collectively a cloud.
A version of Unger’s argument assumes there is just one (typical) instantiation of
cloud in the sky (by the members of s0), and uses PMD5 to conclude that there
are many such (one by the members of each si). The problem therefore remains,
despite the fact that no object is a cloud. Since Compositional Nihilism does not
solve the problem, it is not a problem about the existence of composite objects.
Since the existence of the fusions Di is inessential to Unger’s argument, it will
do no harm to speak as if they do exist, or as if they were candidates to be clouds
(rather than merely constitute clouds), in the remainder. Our discussion can always
be reformulated in terms of plurals and PMD4, in place of fusions and PMD1.
Two Problems 32
1.1.3 Principles of minute differences
We’ve got a version of Unger’s argument in place that doesn’t rely on any contro-
versial assumptions about the existence of objects like the Di’s. The arguments key
premiss is the principle of minute differences PMD4. This section addresses fol-
lowing question of whether principles like PMD1–5 are true. We use ‘PMD’ as a
generic term for all such principles.
We begin by rejecting two arguments against PMD, before turning to two posi-
tive arguments for them. Although these arguments aren’t decisive, they do reveal
that there’s much work to be done before rejection of PMD can provide a satis-
factory response to Unger. The next section then presents a variant on Unger’s
argument that doesn’t require PMD.
1.1.3.1 Two bad arguments against the principle of minute differences
This section dispenses with two bad arguments against PMD. The first is that PMD
is a tolerance principle, and hence known to be false. A tolerance principle for G
has the form:22
If x and y differ minutely w.r.t. F and x is a G, then y is a G.
A PMD for G is not of this form, but rather:
If x and y differ minutely w.r.t. F and x is a typical G, then y is a G.
Say that individuals x1, . . . , xn are a Sorites series for G iff (i) x1 is a paradigm G, (ii)
xn is a paradigm non-G, and (iii) each xi differs only minutely from xi−1 in respects
relevant to G (where 1 < i ≤ n). Since F in a tolerance principle is a respect
relevant to G, a Sorites series for G is a counterexample to a tolerance principle for
G. But a Sorites series for G is not a counterexample to a PMD for G; for a PMD for
G implies that x2 is a G, not that it is a typical-G, and hence implies nothing about
x3–xn. The (possible) existence of a Sorties series for G therefore refutes a tolerance
principle for G without refuting the relevant PMD. Since an appropriate Sorites
series can be constructed for most, if not all, ordinary sortals, those sortals are not
22 Our tolerance principles are a variant on those introduced by Wright (1976).
Two Problems 33
governed by tolerance principles. It remains an open question whether they are
governed by PMD.
We now address the second bad argument against PMD. This argument claims
that there are counterexamples to PMD that are independent of Unger’s puzzle.
Consider the set whose members are our typical cloud C’s constituent water droplets
and one atom of the British Museum. Hud Hudson (2001, p.26) observes that the
fusion of this set resembles C very closely, but is clearly not a cloud. Brian Weather-
son (2009, §7.2) cites this as a counterexample to PMD .23 But this only shows that
a more nuanced understanding of cloud-respects is required. Two objects may be
very similar overall, despite being highly dissimilar in some more specific respect.
The factors relevant to being a cloud, and hence to similarity in cloud-respects, are
weighted: large spatial discontinuities count strongly against being a cloud (resem-
bling in cloud-respects), despite counting for little overall difference.
1.1.3.2 Two justifications for the principle of minute differences
Should we endorse PMD? This section considers two reasons to do so. Although
neither is decisive, they do revel that rejection of PMD is not an easy response to
Unger’s challenge.
The first reason to endorse PMD is that it seems analytic. Ignoring Unger’s puz-
zle, PMD seems beyond reproach. Indeed, it is not merely attractive, but plausibly
partially constitutive of being a typical cloud: how could something that differs
minutely in relevant respects from non-cases be a typical, or paradigm, case? Of
course, the principle’s falsity is compatible with its being intuitively compelling.
Ignoring Russell’s paradox, the naïve comprehension principle is compelling:
For any (possibly complex) predicate F in the language of set-theory, {x : Fx}
exists.
But Russell’s paradox refutes this nonetheless. Maybe Unger’s puzzle refutes PMD.
Still, a solution that retained it would be, ceteris paribus, preferable to one that did
23 Weatherson’s purpose is not Hudson’s. Hudson uses this case to illustrate how something’s
being a cloud is sensitive to otherwise small differences, and hence the sensitivity of cloud-respects
to those differences, whereas Weatherson thinks that small differences in cloud-respects can ground
large differences w.r.t. being a cloud.
Two Problems 34
not. Furthermore, PMD encodes a conception of paradigms at least as compelling
as naïve comprehension for sets.24 If Russell’s paradox is really a paradox, then so
is the Problem of the Many, if giving up PMD is what it requires.
Unger (1980, p.161) suggests a second reason to grant PMD: it follows from the
vagueness of ‘cloud’. According to Unger (1979, §2; 2006a, appendix to ch.7), each
vague concept G obeys a vagueness condition:
For some dimension, or respect, F, sufficiently small differences w.r.t. F can-
not differentiate a G from a non-G.
This is a tolerance principle by another name. Although he is not explicit about
just how this leads to PMD, the idea seems to be this. Were PMD false, then some
typical cloud would differ minutely in cloud-respects from a non-cloud. This is
incompatible with tolerance for clouds, and hence with their vagueness. So clouds
obey PMD.
This is dubious. We’ve already seen that if a Sorites series for G is possible,
then the relevant tolerance principle for G is not a conceptual truth. Since such
a series is typically possible, vague concepts are either incoherent or do not obey
tolerance principles.25 Unger endorses the first disjunct. Since most, if not all,
ordinary concepts are vague, his view is both problematic and highly controversial
(for discussion, see Williamson, 1994, ch.6). So let us set it aside. We should reject
tolerance principles, and with them this second justification for PMD. But this is
no easy way out. Vagueness is paradoxical precisely because tolerance principles
are compelling. Granted that tolerance principles imply PMD, the Problem of the
Many is no less problematic than the Sorites, if rejection of PMD is what it requires.
Both justifications for PMD rest on intuitively compelling claims. (i) PMD is
analytic of ‘typical cloud’. (ii) PMD follows from the tolerance principle underlying
the vagueness of ‘cloud’. The lesson of Unger’s puzzle might well be that these
otherwise attractive theses about typicality and vagueness are false. This is not
to say that solving the puzzle will be easy. Accounts of vagueness and typicality
24 Indeed, PMD probably enjoys considerably stronger intuitive support than does naïve compre-
hension, because set-theory enjoys little or no intuitive support.25 There is a third option: the logic of vagueness is weaker than classical, and even intuitionistic,
logic. We won’t consider this radical view.
Two Problems 35
that violate (i) and (ii) are required, alongside an explanation for why these false
claims seem compelling. However, even rejecting PMD cannot solve the problem.
§10 of Unger’s original article presents an alternative route from many candidates
to many clouds, and the hundred-page discussion in Unger (2006a, ch.7) does not
mention PMD at all. To this we now turn.
1.1.4 Selection and exclusion principles
In §10 of “The problem of the many”, Unger presents a variant on his original
argument that doesn’t rely on an appeal to PMD. This section presents this variant,
followed by three kinds of inadequate response.
1.1.4.1 The alternative argument
Recall the lumps Di that are fusions of the sets si of water droplets in the vicinity
of our cloud C. If C is the only cloud in the sky, then exactly one Di constitutes
a cloud. Which? The alternative version of Unger’s argument is driven by two
difficulties concerning this question. The first concerns answering it. The second
concerns seeing how there could even be an answer.
A selection principle provides a property possessed by exactly one Di, and in
virtue of which it constitutes a cloud. An exclusion principle provides a property
possessed by all bar one Di, and in virtue of which they don’t constitute clouds.
Unless there are such principles, either all or none of the Di’s constitute clouds;
they differ too little in relevant respects for only one to constitute a cloud. In order
for only one candidate to constitute a cloud, a selection principle is required to
privilege it over all others, or an exclusion principle to rule out all candidates other
than it. Unger claims that there are no such principles, and hence that each Di
constitutes a cloud.
If C is the only cloud, then one true selection principle provides the property
of constituting a cloud. And true exclusion principles provide the properties of not
constituting a cloud, and substantially but not totally overlapping a cloud. But these
are obviously either trivial or circular.26 Are there non-trivial and non-circular
26 Circular, in the sense that if x fails to constitute a cloud only because x substantially but not
Two Problems 36
alternatives?
Because of how closely the Di’s resemble one another, the only candidates seem
to involve either the identities of particular candidates, or very fine-grained de-
scriptions of their microphysical make-up.27 The former fails because any lump
that actually constitutes a cloud could have failed to do so. The latter fails be-
cause (i) not all clouds are microphysical duplicates of C, and (ii) some Di that
doesn’t constitute a cloud would have done so had the Dj that actually does so
not existed (because its “extra” droplet hadn’t existed). A selection principle that
accommodated (i) and (ii) would have to be of the form: in conditions C1, lumps
with property F1 constitute clouds; in conditions C2, lumps with property F2 con-
stitute clouds. . . . This is problematic because (a) the constitution of clouds should
turn on general features instantiable in a range of circumstances, and (b) it is hard
to believe that cloud-constitution turns on microphysical structural properties so
fine-grained as to distinguish some Di from all others. Surely reality does not con-
tain substantial distinctions grounded in such slim differences.
We now consider two potential non-trivial and non-circular alternatives: a se-
lection principle in §1.1.4.2, and an exclusion principle in §1.1.4.3. I know of no
alternatives. So §1.1.4.4 examines a position that rejects Unger’s demand for selec-
tion and exclusion principles.
1.1.4.2 Maximality
Sider (2001b, 2003) notes that ordinary sortal properties like cloud are maximal:
“A property, F, is maximal iff, roughly, large parts of an F are not them-
selves Fs.”(Sider, 2001b, p.357)
Large parts of houses are not themselves houses, large parts of people are not them-
selves people, and large parts of clouds are not themselves clouds. Can this solve
Unger’s puzzle? It seems not.
totally overlaps a cloud y, the question is only pushed back to: why is it y, rather than x that’s
disqualified from constituting a cloud? Further selection or exclusion principles are then required.27 These principles will have to be so fine-grained as to distinguish between lumps that differ by
only a pair of water droplets close to their boundaries.
Two Problems 37
Not all ordinary kinds are maximal. Richard Sharvy (1980) considers a table
made by putting two smaller tables together; the smaller tables do not go out of
existence or cease to be tables. Popes have worn crowns (the Papal Tiara) com-
prising two or even three distinct crowns (Wiggins, 1980, p.73). There are two
reasons, however, why we should be reluctant to reject maximality on the basis of
such examples. The first is that it is hard to find counterexamples involving non-
artefactual kinds, and it is these for which Unger’s puzzle is most pressing. The
second is that these are not obviously counterexamples. A counterexample to max-
imality for F is a situation containing a1, . . . , an, all of which are Fs, and where (i)
a1, . . . , an−1 are large parts of an, and (ii) the correct answer to the question “How
many Fs?” is “n”. Sharvy’s table and the Papal Tiara plausibly fail condition (ii).
Asked how many tables there are, Sharvy could answer “One”, or “Two”, but not
“Three” (and certainly not without qualification).28
Recall our original candidates C and D:
Cam bm Dam bmAlthough they almost entirely overlap, neither candidate includes the other. So
maximality disqualifies neither from constituting a cloud, and hence doesn’t solve
the problem.
Maximality will, however, reduce the extent of the problem. Let E be a fusion
of D and E:
Eam bmIf E is a candidate, then maximality excludes both C and D. More generally, if there
is a unique largest candidate—a unique candidate that includes all candidates that
include it—then maximality ensures that only it constitutes a cloud. But there is
no guarantee that there will be a unique largest candidate. There are two ways to
28 Rumfitt (2002) contains relevant discussion.
Two Problems 38
see this.
Firstly, why think that E is a candidate? The fusion of two candidates is not
generally a candidate. What difference does the extent of their overlap make? Why
believe that the fusion of near-coincident candidates will always be a candidate?
Without reason to do so, we lack reason to believe that maximality will narrow the
candidates down to one.
Secondly, suppose E is a largest candidate. Some object F differs from E by
(i) including some droplet just in E’s exterior, and (ii) excluding some droplet on
E’s boundary. Neither E nor F includes the other. Just the same reasoning that
led us to recognise D as a candidate given that C is a candidate, should lead us
to recognise F as a candidate given that E is a candidate; for supposing E to be a
largest candidate is silent about the underlying problem, namely that many nearby
objects are extremely similar in all relevant respects to whichever object possesses
whichever property concerns us.
The point is that, whatever it takes to be a cloud, Unger’s puzzle already con-
cerned it: many objects in the vicinity of each typical cloud resemble it so closely
that it seems arbitrary for just one of them to be (constitute) a cloud. Identifying the
cloud with a fusion of what were previously regarded as cloud-candidates does not
undermine this, but merely changes the topic, diverting attention to a new puzzle.
Even setting aside these concerns about whether maximality reduces the can-
didates to one, it is unclear whether it would solve the problem by doing so. The
issue turns on whether we understand maximality semantically, or metaphysically.
On a semantic construal of maximality, it governs the application of predicates
and concepts: nothing satisfies ‘cloud’ if it is part of something else that does so,
even if it is otherwise just like something that satisfies ‘cloud’. On this reading,
maximality implies extrinsicality.
The metaphysical construal governs the sortal property cloud: nothing that in-
stantiates cloud is part of something else that does so. This is neutral about intrin-
sicality. The property cloud may be maximal because whether an object instantiates
it depends on the object’s external environment. In that case, cloud will be extrin-
sic. But another way in which cloud could be maximal is for the boundaries of its
possessors to vary depending on their external environment; in which case cloud
Two Problems 39
may be intrinsic.29
If Unger’s abundance of near-coincident clouds gives rise to any genuinely meta-
physical problems—i.e. to problems that wouldn’t have arisen had we used words
differently and that can’t be resolved by a more nuanced conception of word-world
relations—then semantic-maximality cannot help, even by reducing the candidates
to one. (For related discussion, see §3.3.1). So far, the only problem concerns rad-
ical error in ordinary numerical judgements. Other problems will be presented
in §1.4. If these are genuinely metaphysical, then no purely semantic techniques
can resolve them. Unlike semantic-maximality, metaphysical-maximality can help
with such problems; but we should doubt whether even that will reduce the candi-
dates to one.
Maximality is guaranteed neither to reduce the candidates to one, nor to solve
the problem even if it did so. So let us set it aside.
1.1.4.3 Massive overlap
Clouds plausibly satisfy:
If x and y massively overlap, then they are not both clouds.
Like maximality, this exclusion principle can be understood either semantically or
metaphysically; the last section’s discussion carries over wholesale.
Most ordinary kinds seem to satisfy this exclusion principle. It implies that at
most one candidate constitutes a cloud. But it provides no reason to think that
any particular one does, in preference to all others. Unless supplemented with a
uniquely satisfied selection principle, it therefore implies that no candidate consti-
tutes a cloud. Massive overlap alone cannot solve the problem.
1.1.4.4 Brutalism
We’ve just seen an unsatisfactory selection principle (maximality) and an unsatis-
factory exclusion principle (massive overlap). It is not clear what other candidates
might be appealed to. So this section examines a view that rejects Unger’s demand
29 Sider (2001b, §1) claims that one adequacy constraint on analyses of intrinsicality is that maxi-
mality implies extrinsicality. He does so because he is assuming a semantic conception of maximality.
Two Problems 40
for selection and exclusion principles. In doing so, we’ll clarify Unger’s argument
and just what that demand amounts to.
Consider (a version of) Peter van Inwagen’s (1990) Special Composition Ques-
tion (SCQ):
SCQ Under what conditions does a set have a fusion?
Ned Markosian (1998) offers the following answer:
Compositional Brutalism There is no true, non-trivial and finitely long answer to
SCQ.
Markosian claims that compositional facts are, in this sense, brute facts. Granted
this, he claims, Unger’s puzzle has an easy “solution”: exactly one set si of droplets
in the vicinity of C has a fusion, this fusion is the only cloud in the sky, and there
is no finitely statable non-trivial reason why this set was selected and all others
excluded.
Compositional Brutalism is not what’s doing the real work here. We saw that
Unger’s puzzle is primarily a puzzle about the instantiation of ordinary kinds, and
only indirectly about composition. To get a puzzle about composition, we need
to assume first that ordinary objects are composite individuals (since the problem
arises even if Compositional Nihilism is true), and second that all reasonably large
composite objects belong to ordinary kinds (since the problem arises even if Unre-
stricted Fusion is true).
To see what’s really doing the work, consider this K-Constitution Question
(KCQ):
KCQ Under what conditions does a set have a fusion that belongs to the ordinary
kind K?
The analogue to Compositional Brutalism is:
K-Brutalism There is no true, non-trivial and finitely long answer to KCQ.
K-Brutalism is compatible with very liberal theses about fusion, but offers a “solu-
tion” to Unger’s puzzle: exactly one set si has a fusion that belongs to the kind cloud,
Two Problems 41
though there is no finitely statable non-trivial reason why that set was selected and
all others excluded.
One objection to both forms of Brutalism is that the relevant facts are law-
governed. Clouds, cats and other ordinary objects do not simply pop in and out
of existence randomly, but in a regular and highly systematic manner. Microscopic
particles have to be appropriately arranged in order for them to compose (or cease
to compose) a cloud. A statement of the laws connecting these arrangements to
the existence of clouds would answer those versions of SCQ and KCQ that concern
clouds.
The objection fails because the Brutalist can consistently grant that cloud-
constitution is law-governed, alongside either of the following theses. (i) The laws
are so complex as to resist finite non-trivial statement. This makes our inability
to state non-trivial selection and exclusion principles a consequence of our epis-
temic, cognitive and practical limitations. (ii) Although finitely and non-trivially
statable, the laws serve only to delimit a class of candidates; the question of which
member of this class constitutes (or composes) a cloud has only a trivial answer.
We consider the Brutalist of kind (ii) before returning to (i).
One might object that thesis (ii) makes it arbitrary which si composes a cloud.
What is the relevant sense of arbitrariness? Suppose that only s0 composes a cloud,
and hence that D0 either is, or constitutes, a cloud. In what sense is this arbi-
trary? The Brutalist posits a significant macroscopic difference between D0 and
each other Di, despite their being no correlated significant microscopic difference.
Why should this be objectionable? Why must all significant distinctions be re-
vealed by microphysical descriptions? Relatedly, we can see the Brutalist as oppos-
ing the demand for an analysis of macroscopic kinds in microscopic terms. Since
we shouldn’t expect any such analysis, the Brutalist may claim, we shouldn’t expect
non-trivial selection and exclusion principles in the first place.
Were the demand for selection and exclusion principles motivated by a demand
for a microscopic analysis of macroscopic phenomena, this would be an effective
reply. But it need not be so motivated. It is better to see Unger as presenting a
challenge to our ordinary world-view: how could only one candidate constitute
a cloud, given their extremely close similarity? On what grounds do we retain
Two Problems 42
our belief in only one cloud when presented with the candidates and their close
similarities? Plausible selection and exclusion principles would provide the best
grounds for doing so. This challenge is not met by simply blocking the argument
from the impossibility of stating non-trivial selection and exclusion principle to an
abundance of clouds. For we may still draw the disjunctive conclusion that either
(a) there are many clouds where we thought there to be one, or (b) there is only
one cloud, though there is no non-trivial reason why it is constituted by, say, D1
rather than D560. An adequate response to Unger’s challenge must provide reason
to endorse disjunct (b) over (a). Brutalism does not.
This exposes the flaw underlying Brutalism of form (i). The Brutalist is surely
right to claim that our inability to find non-trivial selection and exclusion princi-
ples may be a result of our own limitations. But why should we believe that it is?
Brutalism alone provides no reason to do so. So Brutalism fails to address Unger’s
challenge.
Why endorse the Brutalist solution? Perhaps the best reason is due to W.V.O.
Quine (1981b). Quine first endorses three principles: (i) there is at least one cloud
in the sky; (ii) distinct clouds don’t substantially overlap; (iii) material objects are
the material contents of filled spatiotemporal regions. He argues from these to the
Brutalist solution, via:
Bivalence For any statement A, either A is true or A is false.
Here’s Quine’s argument:
“[W]e are committed. . . to treating the table as one and not another of
this multitude of imperceptibly divergent physical objects. Such is bi-
valence.. . . If the term ‘table’ is to be reconciled with bivalence, we must
posit an exact demarcation, exact to the last molecule, even though we
cannot specify it. We must hold that there are physical objects, coin-
cident except for one molecule, such that one is a table and the other
is not.. . . In this way simplicity of theory has been served.. . . [B]ivalence
requires us. . . to view each general term, for example ‘table’, as true or
false of objects even in the absence of what we in our bivalent way are
prepared to recognize as objective fact. At this point, if not before,
Two Problems 43
the creative element in theory-building may be felt to be getting out
of hand, and second thoughts on bivalence may arise.” (Quine, 1981b,
p.36)
Thus the general methodological principles governing theory-choice, notably the
search for simplicity, that motivate Bivalence also motivate the Brutalist solution,
though Quine acknowledges that this is a cost of Bivalence.
Now, the Brutalist solution is certainly a cost. Is it one we should pay? Note
first that simplicity is only one theoretical virtue amongst many. If this kind of
cost-benefit analysis is to motivate the Brualist solution, then the virtues and vices
of its rivals must also be assessed. Those rivals include rejection of Quine’s (i)–
(iii).30 They also include attempts to mitigate the cost of a non-classical semantics
by adopting one that retains classical logic. Supervaluationism is (sometimes pre-
sented as) one such semantic theory. This view is examined in chapter 2, alongside
a variant position that retains both classical logic and semantics. Chapter 3 investi-
gates an application of this variant to the Problem of the Many in a manner conso-
nant with Quine’s (i)–(iii) and without commitment to Brutalism. General method-
ological principles can bring commitment to the Brutalist solution only once the
alternatives have been properly examined, and hence only upon completion of the
investigation we are presently beginning.
1.1.5 Extension to other ordinary kinds
We’ve focused on clouds. Accepting the puzzle’s conclusion—that there are many
where we thought there to be one—wouldn’t be so bad if it arose only for clouds.
This section extends it to all other ordinary kinds of object.
1.1.5.1 Ordinary objects and sorts
Our primary concern in the remainder will be with ordinary objects and the ordi-
nary kinds to which they belong. These objects are the subjects of ordinary talk,
30 Unger’s own response to his puzzle is to reject (i): he takes it so show that our concept of a
cloud is incoherent. Accepting an abundance of clouds involves rejecting (ii). Chapter 4 develops a
rejection of (iii).
Two Problems 44
thought and perception, and the kinds to which those subjects belong. Ordinary
objects are the most commonplace inhabitants of the everyday world around us,
e.g.: grains of sand, bricks, tables and organisms. They also include larger and
smaller objects—e.g. viruses, microbes, planets and galaxies—whose recognition
requires specialist equipment, provided they have similar internal complexity and
coherence to the paradigm cases.
Of all the kinds to which ordinary objects belong, our primary interest is in
sortals. Sortal properties are those, like human, cat and cloud that (by and large)
suffice for countability and demarcate their bearers from other individuals. For
example, although water is a natural kind, the question “How many waters are in
the cup?”, can only be met by mystification and a request for further information:
having counted this water, should that water also be counted? (And how big is
that water anyways?) Sortals like cat and cloud, by contrast, automatically supply
answers to this question.
It does not follow that “How many F’s?” can always be answered when F is
a sortal. It may be vague whether x is a cat, or whether x is a different cat from
y. It may not even be possible for beings with our limited practical and epistemic
capacities to arrive at a settled answer. But these are the exceptions. Ordinary sorts
typically permit counts with determinate correct answers. One lesson of Unger’s
puzzle might be that sortals are more like non-sortals in this respect than it would
otherwise seem.
These characterisations of ordinary objects and sortals are both somewhat im-
precise. But in combination with the paradigms listed, they will suffice for our
purposes. Our concern is not to precisely delimit the notions of sortal and ordi-
nary object, but to show how widespread Unger’s puzzle is.
1.1.5.2 The problem for non-clouds
Unger’s puzzle arises because almost coincident with each typical cloud are many
objects that resemble it extremely closely in cloud-respects. The same applies to
other sorts of ordinary objects. They all suffer the same difficulty: what guarantees
that exactly one candidate will be better-placed than all others nearby?
Two Problems 45
In the case of clouds, the problem results from the gradual decrease in droplet
density across their boundaries. This ensures that inclusion of a single droplet
only just in the cloud’s exterior will be both possible and provide something that
resembles the cloud extremely closely. For most objects however, mere proximity
to their boundaries is insufficient for such close resemblance. The fusion of Tibbles
the cat with a dust particle resting on his skin does not closely resemble Tibbles in
cat-respects.31 In order for an entity to be part of a cat, it entity must be appropri-
ately related to other parts of that cat. Unlike clouds, the appropriate relations are
not purely spatial, but include causal and attachment relations, and maybe oth-
ers besides. Whatever the precise details, let us lump these relations together as
cat-bonding relations:
If x is a cat, then y is part of x iff y is cat-bonded to some part of x.
Causal interaction, attachment, proximity and the like are all matters of degree.
It is thus very plausible that cat-bonding is also a matter of degree. This is fur-
ther supported by the observation that cat’s hairs don’t simply “pop out” between
instants, but gradually work their way loose. Similar remarks apply to all other
ordinary sorts.
The gradualness of cat-bonding suffices to generate Unger’s puzzle. Suppose
that hair h has only just ceased to be part of Tibbles; were h any better cat-bonded
to Tibbles, it would be part of him.32 Let Tigger be a fusion of Tibbles with h.
Tigger resembles Tibbles extremely closely in cat-respects. A principle of minute
differences for cats implies that Tigger is a cat:
If x is a typical cat and y differs only minutely in cat-respects from x, then y
is also a cat.
In addition to this argument via a principle of minute differences, Unger’s second
version of the puzzle also arises. The prospects of stating non-trivial and non-
circular selection and exclusion principles look no better here than in the original
case of clouds. So what guarantees that there is a unique most inclusive cat-like
31 Resemblance in cat-respects is resemblance in those respects relevant to whether something is
(or constitutes) a cat.32 The problem is exaggerated by considering individual molecules, electrons and the like.
Two Problems 46
object on Tibbles’s mat? Without an answer, our belief that there’s only cat on that
mat looks no better off than our belief that there’s only one cloud in the sky. Since
this turns on no peculiarity of cats, the generalisation to all other ordinary sorts is
straightforward.
Another way of seeing the problem is as the challenge of matching each macro-
scopic object with a unique collection of microscopic particles, or infinitely precise
region of spacetime. This seems required for Tibbles to be the only cat on his mat;
for if there are many equally good ways of associating Tibbles with (not entirely co-
incident) collections of microscopic particles, then each such collection is equally
suited to compose a cat; how, then, could only of them do so? The problem is that
our conception of ordinary macroscopic objects doesn’t appear to provide for such
fine-grained distinctions amongst microscopically individuated collections of par-
ticles or lumps of matter: many are ruled out, but more than one remains. Tibbles’s
boundaries are determined by, for example, the attachment of hairs. But unless the
event of Tibbles’s hair falling out admits of uniquely correct re-description as an
event in which such-and-such microscopic entities (and no others) participated,
then the features that determine Tibbles’s boundaries will not suffice to distin-
guish amongst closely resembling lumps (or collections of particles). Until such
a uniquely correct microphysical re-description is supplied, there is no reason to
believe that one is possible; this is all that Unger’s puzzle requires.
The problem is most striking for beings like ourselves. If the argument is sound,
then millions of humans are sitting in your chair right now. A similar argument
shows that if any of these are persons, then so are all the others: millions of people
are sitting in your chair, reading this thesis. Whatever our views about the number
of clouds in the sky or cats on Tibbles’s mat, this is difficult to take seriously.
1.2 Lewis’s puzzle
In Lewis’s hands, Unger’s puzzle becomes a puzzle of vagueness. Here’s his initial
presentation:
“Think of a cloud—just one cloud, and around it clear blue sky. Seen
from the ground, the cloud may seem to have a sharp boundary. Not so.
Two Problems 47
The cloud is a swarm of water droplets. At the outskirts of the cloud,
the density of the droplets falls off. Eventually they are so few and far
between that we may hesitate to say that the outlying droplets are still
part of the cloud at all; perhaps we might better say only that they are
near the cloud. But the transition is gradual. Many surfaces are equally
good candidates to be the boundary of the cloud. Therefore many ag-
gregates of droplets, some more inclusive and some less inclusive, (and
some inclusive in different ways than others), are equally good candi-
dates to be the cloud. Since they have equal claim, how can we say that
the cloud is one of these aggregates rather than another? But if all of
them count as clouds, then we have many clouds rather than one. And
if none of them count, each one being ruled out because of the compe-
tition from the others, then we have no cloud. How is it, then, that we
have just one cloud? And yet we do.” (Lewis, 1993a, p.164)
Later, he gives another:
“Cat Tibbles is alone on the mat. Tibbles has hairs h1, h2, . . . , h1000. Let c
be all of Tibbles including all these hairs; let c1 be all of Tibbles except
for h1; and similarly for c2, . . . , c1000. Each of these c’s is a cat. So in-
stead of one cat on the mat, Tibbles, we have at least 1001 cats—which
is absurd.. . . Why should we think that each cn is a cat?. . . [S]uppose it
is spring, and Tibbles is shedding. When a cat sheds, the hairs do not
come popping off; they become gradually looser, until finally they are
held in place only by the hairs around them. By the end of this gradual
process, the loose hairs are no longer parts of the cat. Sometime before
the end, they are questionable parts: not definitely still parts of the cat,
not definitely not. Suppose each of h1, h2, . . . , h1000 is at this question-
able stage. Now indeed all of c1, c2, . . . , c1000 and also c which includes
all the questionable hairs, have equal claim to be a cat, and equal claim
to be Tibbles. So now we have 1001 cats. (Indeed, we have many more
than that. For instance there is the cat that includes all but the four hairs
h6, h408, h882, and h907.) The paradox of 1001 cats. . . is another instance
Two Problems 48
of Unger’s problem of the many.” (Lewis, 1993a, pp.166–7)
This section examines this argument and its relationship to Unger’s puzzle.
With one exception, Lewis’s first presentation is close to Unger’s. Unger does
not claim that all the candidate surfaces have equally good claim to be the bound-
ary of the original cloud. Unlike Lewis, he assumes that the cloud’s boundary is
settled, and argues for an abundance of equally cloud-like boundaries regardless,
though none bounds the original cloud. Why the difference?
The answer lies in Lewis’s second presentation: the boundaries differ only by
the inclusion and exclusion of “questionable parts”, entities neither definitely part
nor definitely not part of Tibbles. Because of this appeal to questionable parts,
Lewis’s conclusion differs slightly from Unger’s: although there are many cats on
the mat, they all have equally good claim to be Tibbles, they are all questionably
Tibbles. What is the sense of “questionable” and “definite” here? To illuminate
this, we begin with the phenomenon of vagueness.
1.2.1 Borderline parts
This section examines the notion of a questionable, or borderline, part. We begin
with a brief introduction to vagueness generally, before turning to parthood.
1.2.1.1 Vagueness
A theory-neutral characterisation of vagueness is difficult, if not impossible. So we
proceed via paradigm cases and identification of some characteristic features.
Vagueness is the “fuzziness” of the distinction between, for example, the red
and the orange, the tall and the not tall, the intelligent and the unintelligent, or
the chairs and the stools. In a sufficiently well-stocked spectrum from one of these
poles to its pair, there is no sharp transition from one to the other, but a fuzzy
transitional region. On a colour chart, for example, the red zone does not seem to
abruptly terminate when the orange zone begins, but to blur gradually into it.
The cases in this intermediate zone are the borderline cases. When x is a bor-
derline case of F, the appropriate response to the question “Is x (an) F?”, seems
Two Problems 49
to be neither a simple “Yes” nor a simple “No”. This much is a datum.33 Exactly
what the appropriate response should be is a matter of dispute. We introduce the
notion of clarity to characterise the borderline cases. The positive cases outside the
fuzzy region are clearly cases. The negative cases outside that region are clearly not
cases. The borderline cases are neither clearly cases nor clearly not cases. By way
of example, brake lights are clearly red, oranges are clearly orange, and terracotta
pots are borderline red/orange.
One characteristic of vagueness, as opposed to other forms of unclarity, is that
the extent of fuzziness is itself fuzzy. An example: the clearly red zone on a colour
chart does not seem to abruptly terminate when the borderline red/orange zone
begins, but to blur gradually into it. Fuzziness permeates vague classification. This
gives rise to the phenomenon of higher-order vagueness: borderline cases to the
borderline and clear cases (and borderline cases to those cases etc.). More on this
in §2.9.
Vagueness is also responsible for the Sorites paradox. Consider the claims:
R1 Ten seconds after his birth, Bertrand Russell was young.
R2 Forty years after his birth, Russell was not young.
R3 If Russell was young i seconds after his birth, then he was also young i + 1
seconds after his birth.
All three are intuitively compelling. Indeed, each is plausibly partially constitutive
of the meaning of ‘young’. But they are classically inconsistent.34 For instantiat-
ing R3 for i = 10 followed by modus ponens using R1 leads to the conclusion that
Russell was young eleven seconds after his birth. Instantiating R3 for i = 11 and
another application of modus ponens gives the conclusion that Russell was young
twelve seconds after his birth. Repeating this process eventually gives the conclu-
sion that Russell was young forty years after his birth, which contradicts R2. This
33 I do not claim that neither “Yes” nor “No” may be offered in response the question “Is x (an) F?”
when x is a borderline F. I claim only that such answers should be qualified, to avoid misleading
one’s audience.34 More precisely: R1–R3 are classically inconsistent given some elementary arithmetic and obvi-
ous truths about the structure of time.
Two Problems 50
is an instance of the Sorites paradox. The following claims provide another:
T1 Anyone two-hundred centimetres in height is tall.
T2 Anyone one-hundred centimetres in height is not tall.
T3 If anyone of i centimetres in height is tall, then anyone of i− 1 centimetres in
height is tall.
Again, repeated instantiations of T3 and applications of modus ponens lead from T1
to the conclusion that anyone one-hundred centimetres in height is tall, which con-
tradicts T2. Premisses R3 and T3 express what §1.1.3 called tolerance principles.
We will call them Sorites principles. Their plausibility results from the vagueness of
the property or concept in question, being young in our first case, and being tall in
the second.
The Sorites paradox and borderline cases seem to be connected: the presence
of the borderline cases seems to explain the plausibility of Sorites principles. Con-
sider the negation of R3:
¬∀i(Russell was young i seconds after his birth →
Russell was young i + 1 seconds after his birth)35
This is classically equivalent to:
(1) ∃i(Russell was young i seconds after his birth ∧
¬Russell was young i + 1 seconds after his birth)
Which, if true, has a true instantiation:
Russell was young n seconds after his birth ∧
¬Russell was young n + 1 seconds after his birth
What might the cut-off point n be? Since a conjunction entails its conjuncts, we
should propose only those answers both of whose conjuncts we are prepared to
endorse. There are three cases.35 ‘→’ is the material conditional throughout.
Two Problems 51
First case: Russell was clearly young n seconds after his birth. Then Russell was
either clearly young or borderline young n + 1 seconds after his birth. Either way,
(unqualified) endorsement of ‘Russell was not young n + 1 seconds after his birth’
is inappropriate. So we should not propose n as the cut-off point.
Second case: Russell was borderline young n seconds after his birth. Then (un-
qualified) endorsement of ‘Russell was young n seconds after his birth’ is inappro-
priate. So we should not propose n as the cut-off point.
Third case: Russell was clearly not young n seconds after his birth. Then en-
dorsement of ‘Russell was young n second after his birth’ is clearly inappropriate.
So we should not propose n as the cut-off point.
In each case, we should not propose n as the cut-off point. So we should not
propose n as the cut-off point. Since n was arbitrary: we should not propose, of
any i, that i seconds after his birth is the cut-off point for Russell’s youth. We’ve
just seen that this follows from the following two factors: (i) outright endorsement,
without qualification, of a borderline statement is inappropriate; (ii) the clear pos-
itive and negative cases are separated by borderline cases. The explanation for the
attraction of R3 is then that we reject (1) because we know a priori that we ought not
to endorse any of its instantiations; and since we reject (1), we endorse it’s negation
R3.
This explanation won’t do as it stands; for we accept some existential generali-
sations despite knowing a priori that we ought never to endorse any of their instan-
tiations. Two examples: ‘some mammal was the first unnamed dog born at sea’,
and ‘something very strange happens inside a black-hole’s event-horizon’. Both are
relevantly disanalogous to vagueness, and hence don’t undermine our explanation
for the plausibility of T3. The first turns on semantic vocabulary appearing within
the scope of a quantifier, yet none appears in (1). The second turns on our own
epistemic limitations, but our ignorance of when Russell ceased to be young does
not: it seems, in some sense, misguided even to wonder about, never mind set about
trying to discover, when Russell ceased to be young.36
36 Not all will grant this. Epistemicists like Timothy Williamson claim that we are ignorant of
when the last second of Russell’s youth was (Williamson, 1994). But since not all ignorance results in
vagueness, they too must find a disanalogy between typical ignorance and vagueness. The epistemi-
Two Problems 52
With this preliminary introduction to vagueness complete, we return to Lewis’s
puzzle.
1.2.1.2 Mereological vagueness
Suppose Tibbles is moulting. Let h1, . . . , hn be a series of hairs where (i) h1 is clearly
part of Tibbles, (ii) hn is clearly not part of Tibbles, and (iii) each hi is only very
slightly less firmly attached to Tibbles than hi−1 (where 1 < i ≤ n). h1, . . . , hn are a
Sorites series for being part of Tibbles.
This series seems to exhibit the fuzziness characteristic of vague classification.
The hairs that are part of Tibbles don’t seem to be immediately succeeded in the
series by those that aren’t. Rather, some hairs hm–hm′ are borderline parts of Tib-
bles, separating the clearly attached hairs h1–hm−1 from the clearly detached hairs
hm′+1–hn.37 That there are such borderline hairs is reinforced by the plausibility of
the Sorites principle:
If hi is part of Tibbles, then so is hi+1.
For if some hairs are borderline parts of Tibbles, then the previous section’s expla-
nation for the attraction of Sorites principles generally, can also be used to explain
the attraction of the particular principle above.
Each borderline hair provides a borderline cat-candidate (or cat-constituting-
candidate). Let T be all of Tibbles, excluding h1, . . . , hn; let each Ti amongst T1, . . . , Tn
be a fusion of {T, h1, . . . , hi}. When it’s borderline whether hi is part of Tibbles, it’s
also borderline whether Ti is (constitutes) a cat.
1.2.2 Why many cats?
We now have a range of cat-candidates, namely each fusion Ti of the set {T, h1, . . . , hi}
where hi is a borderline part of Tibbles. Since it’s borderline whether hi is part of
a cat, it’s borderline whether Ti is a cat.38 This much is unproblematic. The prob-
cist may then use this disanalogy to defend the argument from non-endorsement of each instantiation
of (1) to acceptance of T3.37 It will, of course, be a vague matter just where in the series hm and hm′ are located38 We talk for simplicity as if the fusions could be cats, rather than merely constitute cats. We’ve
seen that nothing turns on this.
Two Problems 53
lem arises because Lewis draws the stronger conclusion: each candidate Ti is a cat.
What licenses this?
Suppose that some candidate is more cat-like than any other. Then the others
are surely not borderline cats, but clearly non-cats: the best candidate wins. Since
they are all borderline, all are equally cat-like.39 But one way for one to be more
cat-like than any other is for only it to be a cat. Then since at least one of the
candidates is a cat, they all must be. So there are many cats on Tibbles’s mat.
This argument licenses the following stronger conclusion than Unger’s: for each
candidate Ti, it’s borderline whether Ti is Tibbles, the cat with which we began.
Since each candidate has equal claim to be Tibbles, the result is borderline iden-
tity sentences ‘Ti = Tibbles’.40 This shows that something must have gone wrong
somewhere in Lewis’s argument; for an argument exactly parallel from the candi-
dates all being borderline cats to their all being cats can now be used to show that
they are all Tibbles. Yet that’s impossible because the candidates are many, while
Tibbles is one.
What has gone wrong? The answer must be that although the candidates are all
on a par w.r.t. being cats and it’s clear that one of them is a cat, it doesn’t follow that
any one of them is clearly a cat; they may all be only borderline cats instead. Let
the sentential operator ‘∆’ formalise ‘It is clearly the case that. . . ’. The following
argument-form must be invalid:
∆∃xFx, therefore: ∃x∆Fx.
And the following must be consistent:
∆∃xFx ∧ ∀y(¬∆Fy ∧ ¬∆¬Fy)
This is a constraint on theories of vagueness if (i) the candidates are all borderline
cats, (ii) it’s clear that one of them is a cat, and (iii) it’s clear that there is only one
cat the mat. The supervaluationist views considered later all respect this.
39 Indeed, that the Ti’s are all equally cat-like is what seems to be responsible for their being bor-
derline cases in the first place.40 This doesn’t conflict with Gareth Evans’s (1978) famous argument against vague identity. For an
identity sentence can be vague due to vagueness about the referents of the terms flanking ‘=’, rather
than vagueness in the identity relation. See Lewis (1988b) for discussion.
Two Problems 54
1.3 How many puzzles?
Are Unger’s and Lewis’s puzzles one and the same? It is too early to give a firm
answer, but here are some of the issues.
On the assumption that cat is maximal, Lewis’s puzzle results in more cats than
Unger’s. Each Lewis-candidate Ti includes each candidate Tj when i ≤ j. So max-
imality implies that at most one Lewis-candidate is a cat: when some candidates
are otherwise equally suited to be cats and one includes all others, maximality im-
plies that that largest one is the best. So, ignoring vagueness, there is a unique
largest best candidate in the series T1, . . . , Tn. Hence Unger can recognise only this
one as a cat. Unger’s cats extensively overlap, but don’t include one another. Vague-
ness, however, undermines the thought that there will be a unique best candidate in
the series of increasingly inclusive candidates: several candidates can have equally
good claim to best despite one being clearly largest because it may be borderline
whether a smaller candidate is more cat-like than some larger one.
Another difference concerns the conclusions of the puzzles. As we saw in the
previous section, Lewis’s cats, unlike Unger’s, are not clearly distinct from the cat
with which we began.
Thus we have two reasons to distinguish the puzzles. Unger’s most recent
work on the topic argues for a third: his would arise even were there no vague-
ness (Unger, 2006a, pp.369–70, 468–9, chs.7.8–7.9). Unger claims that there could
be many objects on the mat that differ minutely in cat-respects from Tibbles, even
were it entirely precise and determinate which of these objects coincides with Tib-
bles. So Unger’s puzzle, unlike Lewis’s, does not make essential appeal to vague-
ness in Tibbles’s boundaries.
This argument is only sound if its premiss is true. Is it really possible for
Unger’s puzzle to arise on the assumption that Tibbles is sharply bounded? Not
if Unger’s puzzle is a source of vagueness, or if both puzzles are manifestations of
a single underlying phenomenon. Thus whether there are two puzzles here or one
turns ultimately on the nature of vagueness. It cannot be settled, as Unger wishes,
in isolation from an investigation into the nature of vagueness. Having settled on
an account of vagueness in chapter 2, chapter 3 examines and rejects an applica-
Two Problems 55
tion of it on which Unger’s attempt to distinguish the puzzles is unsound; chapter
4 then endorses an application on which Unger’s attempt succeeds.
One way to see the issue is this. Unger’s is the puzzle of too many candidates:
what guarantees that one candidate is better than all others? If there is no such
guarantee, how can there be only one cat? Lewis’s is the puzzle of borderline
candidates: how can there be only one cat when all the candidates are borderline
(and hence equally good candidates), and yet one of them clearly is/constitutes a
cat? Several questions arise about the relations between these puzzles. Does an
overabundance of best candidates imply the existence of borderline candidates?
Do the existence of borderline candidates imply an overabundance of best candi-
dates? And supposing that the answer to both questions is “Yes”, are these puzzles
both manifestations of a single phenomenon, or of distinct yet mutually entail-
ing phenomena? These questions cannot be settled in isolation from an account
of vagueness. If it turns out that both puzzles are manifestations of a single phe-
nomenon, then the first two reasons to distinguish the puzzles fail: although Unger
may recognise fewer candidates and draw a weaker conclusion than Lewis, that’s
only because he’s failed to recognise the true nature of his puzzle.
1.4 Puzzle or problem?
Why not just accept the conclusion that there are many cats on Tibbles’s mat? It
conflicts with our ordinary world-view, but that’s insufficient for rejection if phi-
losophy can make genuine discoveries. (And what’s the point of philosophy if it
can’t?) This section presents several alternative problems. We do not claim that
any is decisive, only that together they make a cumulative case against accepting
an abundance of cats on Tibbles’s mat.
1.4.1 Time, modality and coincidence
The Problem of the Many is a source of fission and fusion puzzles. Suppose that
Tibbles’s boundaries are entirely precise at time t, and there are no other nearby
Two Problems 56
cat-candidates; neither Unger’s nor Lewis’s puzzle arises at t.41 Suppose also that
Tibbles’s boundaries are extremely vague by the later time t′, so that millions of
almost coincident cats are then on the mat. Which later cats were on the mat at
t? Tibbles seems to have undergone fission. And if Tibbles’s boundaries later be-
come less vague, then those millions of cats seem to undergo fusion. We’ll focus
primarily on fission, but similar considerations arise in both cases.
Fission cases are often presented as a source of insight about the nature of per-
sistence. They are usually thought to be atypical. But if the reasoning behind the
Problem of the Many is sound, then fission is the norm for all kinds of ordinary ob-
ject, including ourselves. Our view about fission had better not, therefore, conflict
with our ordinary judgements of persistence. For example, since cats and persons
survive for extended periods of time, this rules out approaches on which fission
destroys objects, replacing them with two new individuals; yet that’s quite an at-
tractive approach to fission.
Relatedly, fission resulting from the Problem of the Many is unlike “ordinary”
fission. Surely cats are not brought into being by hairs working loose from other
cats; that’s just the wrong kind of change to create a cat. So all the cats on the mat
after fission—following an increase in the extent of the Problem of the Many—were
on the mat beforehand. We now have an argument not just for the near-coincidence
of millions of cats on the mat, but for their (temporary) total coincidence prior to
fission.
A modal analogue strengthens this from temporary coincidence to permanent
coincidence. Suppose our original fission case occurred in world w. Let v be a world
just like w until immediately before Tibbles’s boundaries became vague, when Tib-
bles was destroyed. How many cats are on Tibbles’s mat in v? Surely there should
be only one; there is no vagueness in the boundaries of any cat on the mat in v. But
since cats don’t come into existence when hairs fall out of other cats, all the post-
fission cats in w must be present in v too, where they coincide throughout their
entire lives.41 The assumption that Tibbles’s boundaries were ever utterly precise is inessential; only the
weaker assumption that the extent of their vagueness can vary over time is necessary. The stronger
assumption simplifies presentation.
Two Problems 57
The alternative, that there is only one cat on the mat in v, seems to imply that
the number of earlier cats depends on the future course of events. For v is just
like an initial segment of w, except only that there are fewer cats in v than in w.
Yet surely the number of past cats shouldn’t depend on how things turn out in the
future.
Something similar applies even if every cat in w is also in v. What reason is
there, in v, for the existence of all those cats? There seems to be none, other than
to accommodate the (merely) possible future extent of vagueness. Yet surely the
number of past and present cats should not depend on how events could (and in
this case didn’t) unfold in the future. This approach will also increase the extent
of total coincidence: there are as many cats in v as there are in any possible future
that could have unfolded from some time in v. Since Tibbles’s boundaries could
have, but didn’t, become very vague indeed, very many cats are on the mat in v and
coincident throughout their entire lives; there is no reason in v for any of these cats
to exist other than that events could have (but didn’t) unfolded so that they didn’t
quite coincide with Tibbles.
The near-coincidence of cats in our original version of the problem might not
concern us. After all, partially overlapping objects are commonplace. But we have
just argued for the permanent total coincidence of Tibbles with millions of cats in
v. Even defenders of coincident entities might balk at this. David Wiggins (1968),
for example, grants that distinct objects can, and often do, occupy the same place at
the same time, but denies that objects of the same kind can do so even momentarily.
Can we avoid these coincident cats in v? Chris Hughes (1986) surveys the op-
tions. First option: our description of v is multiply satisfied; really there are mil-
lions of worlds qualitatively just like v, each of which contains only one cat, a dif-
ferent one in each. One consequence is that a world’s qualitative history plus the
identities of everything bar the cat(s) on the mat, is insufficient to determine which
cat is on the mat. It is also hard to regard these worlds as genuinely possible. Pre-
sumably, any cat would still exist were its boundaries a little more precise. Yet on a
standard possible-worlds style semantics for counterfactuals, the current proposal
will falsify this.42
42 On the Lewis-Stalnaker semantics: pA�→ Cq is true at w iff every closest world to w at which
Two Problems 58
Second option: our concept cat is not a single concept; millions of different cat-
concepts determine just slightly different cat-like paths through modal space. Each
cat-concept pairs one object on the mat in w with one on the mat in v, and applies
to no more than one such in each. This view replaces our natural kind concept
cat with many different such concepts. So the candidates don’t all belong to the
same kind. Yet surely there are not so many natural kinds. And how could these
coincident objects that come into existence as a result of the same natural processes
and which are capable of breeding with exactly the same objects really belong to
different kinds? This abundance of biological kinds is not even discoverable by the
standard methods of biologists, but only by a priori reflection on the boundaries of
cats.
Only this second option is available in the purely temporal version of the prob-
lem. There is no space to claim that our descriptions of the past fail to distinguish
between an array of qualitatively similar pasts in which different objects exist. So
we seem committed to either (a) the dependence of the past on the future and the
coincidence of objects of a kind, or (b) implausible differences in kind between the
objects on the mat.
1.4.2 Causation
Trenton Merricks (2003) presents the following argument against the existence of
most ordinary objects:
(i) If ordinary objects exist, then they cause the same effects as their constituent
atoms acting in concert.
(ii) If ordinary objects cause the same effects as their constituent atoms acting in
concert, then there is widespread and systematic causal overdetermination.
(iii) There is no widespread and systematic causal overdetermination.
(iv) So there are no ordinary objects.
A is true is a world at which C is true. On the view described in the text, no candidate exists in each
closest world to w in which the cat on the mat’s boundaries are precise. So the following is untrue
in w: ‘if Tibbles’s boundaries were a little more precise, then any cat on the mat would still have
existed’.
Two Problems 59
Merricks goes on to maintain that only objects whose causal powers go beyond
those of their constituents avoid elimination via this argument. He also claims that
only conscious objects have this feature, and hence that the only ordinary objects
are conscious objects.
This argument is valid. To resist it, we must resist its premisses. Which?
Consider premiss (iii). Why deny that there is systematic and widespread causal
overdetermination? Though Merricks offers no argument, some will grant it nonethe-
less. The argument’s soundness then turns on premisses (i) and (ii). Yet the Prob-
lem of the Many shows that only premiss (iii) is required.
Suppose that many cats almost coincide with Tibbles. Almost anything that any
one of these cats causes is also caused by each of the others. So if the argument for
many cats is sound, then there is widespread and systematic causal overdetermi-
nation. So by premiss (iii) alone: objects susceptible to the Problem of the Many do
not exist.
1.4.3 Free will
Hudson (2001, ch.1.5) argues that the Problem of the Many challenges our ability
to act freely. Suppose that you are an ordinary material object, a human being
say. By the Problem of the Many: there are many humans in your chair. A similar
argument shows that since you are a person, so are they: many people are sitting in
your chair. Suppose you freely lift your arm. It follows of necessity that each other
person in the chair lifts their arm. The following principle then implies that only
your action was free:
“If (i) A’s freely doing x at t entails B’s doing y at t, and (ii) A freely
does x at t, and (iii) A is distinct from B, then B does not freely do y at
t.” (Hudson, 2001, p.40)
The formulation needs modifying to allow for God’s freedom: your freely lifting
you arm entails God’s letting you do so, but God freely let you do so. We might also
question the application of entailment to actions rather than propositions. Still,
something along these lines is intuitively very plausible.
Two Problems 60
It follows that at most one person in your chair acts freely at any time. Which?
The reasoning behind the original problem leads to the conclusion that either all or
none of them ever acts freely. How could any one be non-arbitrarily selected and
all others excluded from acting freely? Any principled selection and exclusion for
free action would presumably also suffice to ensure that there is only one person or
human in your chair. Since there are many persons in your chair, either all or none
of them ever acts freely. By Hudson’s principle: at most one does. So none do.
1.4.4 Real choice
Unger (2006a, ch.7) argues that your power to make real choices, choices indepen-
dent of those of any other person, shows that you are the only person in your chair.
Suppose you have never previously considered either the concept of a blue
sphere or the concept of a red cube, and also lack pre-existing inclination towards
imagining instances of either concept. (Substitute as required to make this sup-
position true.) Imagine either a blue sphere or a red cube. Write down which
you imagined. Repeat as often as you like. Each time, I assume, you write down
whichever you imagined. Unger thinks this counts against the existence of many
other people in your chair; for were we to ask millions of people to carry out this
experiment, we would expect divergence in their answers. Beings with the power
to make genuine choices will tend to make different choices when they lack prior
inclination towards one option over another.
Unger consider three responses. First response: we lack the power to make
genuine choices; our ability to choose is constrained by the other people we al-
most coincide with. Second response: it’s just pure luck that you and your many
always make the same choice, that they never thwart your decisions. Third re-
sponse: beings with the power to make genuine choices are simple non-physical
entities—hence not susceptible to the Problem of the Many—that causally interact
with their many human bodies. This is Unger’s preferred solution. He misses a
fourth response: our power to make genuine choices is constrained by our physical
make-up in such a way that near-coincident choosers cannot manifest this power
in different ways. On this view, the argument rests on an inadequate metaphysics
Two Problems 61
of choice. This last is perhaps the most attractive option, but each will be objec-
tionable to some.
1.4.5 Moorean fact
Moorean facts are theses whose plausibility is so great that no philosophical argu-
ment could refute them. For each Moorean fact M, it is supposedly more plausible
that any argument (or collection of arguments) against M involves a false premiss
or invalid inference, than that M is false. When we cannot locate this false premiss
or invalid inference, it is supposedly more plausible that we are in error than that
the argument is sound.
Is it a Moorean fact that, sometimes, only one typical cloud is in the sky? It is
certainly very plausible. But belief in the falsity of a Moorean fact is supposed to
be so radical, the departure from our ordinary world view so great, that it cannot
be seriously entertained for long (or outside the philosophy room). An abundance
of clouds where we thought there to be just one does not seem to be of this kind, or
to be nearly so radical as the falsity of standard examples, e.g.: there is an external
world; I have two hands; 2 + 2 = 4; murder is wrong.
That Tibbles is the only cat on his mat looks like a better candidate. But is it
really impossible to believe otherwise? That you are the only person reading this
page looks like a better candidate; it is very strange to think that “you” are not
one person but many. But it’s still not clear that I can’t seriously entertain that
thought. On the other hand, there is surely only one person in my chair. Nobody
else is perceiving my computer screen, considering what I should write next, or
wondering what I’ll have for lunch. That I am the only person in my chair seems
as good a candidate Moorean fact as any. And likewise, mutatis mutandis, for you, I
suppose.
1.4.6 Responsibility
The Problem of the Many threatens our most commonplace methods of apportion-
ing praise, blame and moral responsibility. Suppose that someone commits mur-
der and is punished with a life sentence. The punishment is out of proportion to
Two Problems 62
the crime: the murderer killed millions of people. Furthermore, if Hudson and
Unger’s arguments about freewill and choice are sound, then at most one person
almost coincident with all those who were punished freely chose to commit the
crime. Since one can be justly punished only for what one freely chooses to do,
millions are routinely unjustly punished by even the most careful judicial system.
Similarly, even the smallest charitable donation can improve the lives of thousands,
and monogamy is impossible. It is unclear whether our ordinary moral beliefs and
practices can be reconciled with the Problem of the Many for persons.
1.4.7 Singular thought and reference
The Problem of the Many threatens the possibility of singular, or de re, thought
about particular objects (Unger, 1980, §12A). Suppose Rosie tries to think about a
particular book on her desk. No feature of her thought or perception of the book
privileges just one of the many with which it almost coincides; nothing about Rosie
or the books could disqualify all but one from being the subject of her thought.
In what sense is Rosie’s thought singular? How can Rosie have a singular thought
about a book unless there is some book her thought is about? She surely isn’t having
a different de re thought about each book. (This very last claim is questioned in
§3.2.3).
A more theoretically loaded problem assumes that de re thought is object--
dependent: the possibility of having the thought depends upon the existence of
the particular object it is a thought about. Consider Rosie’s singular thought about
the book on her desk. Let w be a world that differs from actuality only in that one
of the book-candidates does not exist in w. Surely the character of Rosie’s thought
in w is just as it actually is; for all the other candidates exist, and nothing in her re-
lations to the books distinguishes that particular candidate from all others. Rosie’s
singular thought is not dependent on that candidate. Generalising: Rosie’s singular
thought is not dependent on any candidate. So Rosie’s singular thought about the
book is not object-dependent. At the very least, an alternative theoretical charac-
terisation of singular thought is required.
Two Problems 63
1.5 Conclusion
This chapter presented both Unger’s and Lewis’s versions of the Problem of the
Many. We began with two versions of Unger’s argument. The first is a positive
argument from a principle of minute differences to many cats on Tibbles’s mat.
We saw how to formulate this with very weak ontological assumptions, and that
the puzzle is not primarily about the existence of individuals, but the instantiation
of ordinary sortal properties. The second version is best seen as a challenge to our
ordinary belief in just one cat on Tibbles’s mat: what ensures that each macroscopic
object is correlated with a unique class of microscopic constituents? Given this
second version, rejection of the principles of minute differences doesn’t solve the
problem. We then turned to Lewis’s puzzle. This proceeds by appeal to vagueness
and borderline cases of parthood. Again, no controversial assumptions seemed to
be required. Once Unger’s puzzle of too many candidates and Lewis’s puzzle of
borderline candidates were in place, we saw that the question of whether these are
two puzzles or one cannot be settled in isolation from a theory of vagueness.
Both these puzzles seem to arise for all sorts of ordinary macroscopic object,
including ourselves. So we closed with a range of more and less theoretical reasons
to be dissatisfied with simply accepting the conclusion that there are many people
sitting in each of our chairs. The next chapter develops a supervaluationist account
of vagueness. The final two chapters apply this account to the Problems of the
Many in two different ways.
64
Chapter 2
Supervaluationist Theories of
Vagueness
This chapter develops a supervaluationist approach to vagueness. The following
two chapters present different applications of this approach to Unger’s and Lewis’s
puzzles.
Although prominent in the literature on vagueness, supervaluationism is not
a unified theory of vagueness. It is, rather, a collection of views united by a for-
mal framework and the importance of the concept of super-truth to the analysis of
clarity. The three key theses are:
(i) Vague classification is best represented by a class of sharpenings.
(ii) The apparatus of truth-evaluation privileges no sharpening over any other.
(iii) Clear truth is best represented by supertruth, and clear falsity by superfalsity.
§2.1 describes the formal setting and key concepts. §2.2 begins to provide the for-
malism with a philosophical interpretation. A range of interpretations are assessed
in §§2.3–2.4, where all bar two are rejected. These remainders are the focus of the
rest of the chapter. §2.5 presents supervaluationism’s most attractive features. We
turn to a range of objections in §§2.6–2.9. One view will emerge as clearly prefer-
able to the other. Subsequent chapters apply this view to Unger’s and Lewis’s puz-
zles.
Supervaluations 65
Two questions before we begin: (a) why supervaluationism?; (b) why only su-
pervaluationism? In response to (a): because supervaluationism is popular, and
maybe even the standard approach to vagueness insofar as there is such a thing.
In response to (b): because there simply isn’t space here to examine more than one
approach to vagueness in the detail it deserves.
2.1 Supervaluationist formal theory
This section outlines the supervaluationist formalism. The classic presentation is
Kit Fine’s (1975). Because one of Fine’s primary goals is to survey the formal terrain,
his discussion contains more complexity than we require. So we simplify where
possible. In particular, we consider only complete sharpenings that assign a truth-
value to each wff, and not also partial sharpenings that relax this constraint.
Our object-language has the form of standard predicate calculus with identity.
We use the following metalinguistic variables (alongside subscripted, primed and
starred variants): ‘α’ ranges over object-language terms and variables; ‘Φ’ ranges
over object-language predicates (sometimes superscripted to indicate the number
of argument positions); ‘A’, ‘B’ and ‘C’ range over object-language wffs; ‘s’, ‘t’
range over sharpenings; ‘v’ ranges over variable assignments. We won’t always
mark use/mention distinctions, and will sometimes use metalinguistic variables
schematically. Context should make things clear enough.
A sharpening s is a pair 〈Ds, JKs〉. Ds is a set of individuals, the domain of s.1 JKs
is a valuation function from object-language constant terms and predicates such
that:
For each term α : JαKs ∈ Ds.
For each n-place predicate Φn : JΦnKs ⊆ Dns .2
J=Ks = {〈x, x〉 : x ∈ Ds}.1 We call the domain of s the s-domain. Similarly, a predicate’s extension at s is its s-extension, a
sentence true at s is s-true, and so on.2 For simplicity, we often write as if the members of JΦ1Ks were elements of Ds, rather than their
singletons.
Supervaluations 66
By way of initial gloss, JΦKs is the extension of Φ at s: the set of things of which
Φ is s-true. Some interpretations of the formalism mandate revisions to this gloss.
The last of these conditions ensures that identity is classical.
A supervaluationist model M is a class of sharpenings such that:
For any s, t ∈ M and singular term α : JαKs = JαKt.
For any s, t ∈ M : Ds = Dt.
The first condition ensures that singular terms are not a source of vagueness. This
is relaxed in the next chapter (§3.1.1). The second condition ensures that quantifi-
cation is not a source of vagueness.
A variable assignment v is a function from object-language variables α such that:
v(α) ∈ Ds.
Let JKs,v be the function from object-language terms and variables such that:
For each term α : JαKs,v = JαKs.
For each variable α : JαKs,v = v(α).
JαKs,v is the value of α given (i) the assignment JKs of values to constant terms, and
(ii) the assignment v of values to variables.
We use this to recursively define a relation between assignments v, sharpen-
ings s, models M and wffs A, written ‘v, s, M A’. When v, s, M A, we say that
A is true at v, s, M, or that v, s, M makes A true. We say that v, s, M makes A false
iff v, s, M makes ¬A true. Given the clause for ¬ below: v, s, M makes A false iff
v, s, M does not make A true. Thus we can speak of the truth-value of a wff relative
to a model, sharpening and assignment (though reference to a model will often be
left tacit). Although the relativisation to models is inert in the definitions below,
it’s needed to introduce a clarity operator ∆ later. We treat ∨,→, ∃ as defined from
¬,∧, ∀ in the standard way. The base clauses of the definition are:
v, s, M Φnα1, . . . , αn iff 〈Jα1Ks,v, . . . , JαnKs,v〉 ∈ JΦnKs.
v, s, M ¬A iff v, s, M 1 A.
Supervaluations 67
v, s, M A ∧ B iff v, s, M A and v, s, M B.
v, s, M ∀xA iff v′, s, M A, for every assignment v′ that differs from v at
most over ‘x’.
Formally, sharpenings are classical models.
We now drop the relativisation to assignments:
s, M A iff v, s, M A, for all assignments v.
Then we define supertruth and superfalsity in a model:
A is supertrue in M iff, for any sharpening s ∈ M : s, M A.
A is superfalse in M iff, for any sharpening s ∈ M : s, M 1 A.
A is supertrue (superfalse) in M iff every sharpening in M makes A true (false).
Thus we can talk of the supertruth-value of a wff in a model.
Now the formalism is in place, let’s apply it to vagueness.
2.2 Understanding the supervaluationist formalism
The intuitive inspiration for supervaluationism is the idea that vague predicates
can be made precise in many different ways; hence the interest in classes of sharp-
enings. If a predicate F applies to an object o regardless of how F is made precise,
then o is clearly F. If F never applies to o, regardless of how F is made precise, then
o is clearly not F. And if F applies to o under only some ways of making F precise,
then o is borderline F. Hence the identification of clear truth with supertruth and
clear falsity with superfalsity.
This is only a sketch of a guiding picture. What exactly is the sense in which a
vague predicate can be made precise? What is a sharpening? These questions are
best addressed in the context of attempts to delimit a consequence relation under-
lying vague discourse. There are two reasons for this. Firstly, focus on consequence
pins down the relevant notions of content and semantics, thereby helping to elim-
inate terminological disputes: our interest is in logically relevant content. We’ll
see that this constrains permissible accounts of sharpenings and supervaluationist
Supervaluations 68
models. Secondly, because of the Sorites, an account of good deductive inference
within a vague language is arguably the most pressing demand on any theory of
vagueness; and a language’s consequence relation provides the standard of correct-
ness for deductive inferences within it. Thus before we can begin with a philosoph-
ical account of the formalism, an account of the relationship between model-theory
and consequence is required. This is our next topic. Once this account is in place,
§§2.3–2.4 use it to develop two kinds of supervaluationism. These are evaluated in
the rest of the chapter.
2.2.1 Consequence, truth and interpretations
Alfred Tarski (1936) offered the following analysis of consequence:
C is a consequence of Γ iff, for any interpretation I, if every member of Γ is
true under I, then C is true under I.3
Throughout, ‘Γ’ ranges over sets of sentences. An interpretation is an assignment of
logically relevant content to linguistic items, a possible semantic structure. So on
the Tarskian view, the members of Γ jointly imply C iff there’s no way of assigning
content to the members of Γ and to C that makes the former true without also
making the latter true.4
John Etchemendy (1990) criticises Tarski. Here are two examples. (i) were there
only one thing, ‘∃x∃yx 6= y’ would be true on no interpretation and hence, by
Tarski’s lights, be logically false. (ii) In higher-order languages, either the Contin-
uum Hypothesis or its negation is true on all interpretations and hence logically
3 It’s debatable whether this is quite what Tarski intended. It’s certainly how Etchemendy (1990)
interprets him. But Tarski’s own presentation is in terms of models, i.e. mathematical structures,
rather than interpretations. Plausibly however, Tarksi’s models were intended as mathematical rep-
resentations of interpretations.4 Restrictions on which expressions get re-interpreted are required; for A ∧ B wouldn’t imply A
if interpretations of ∧ as disjunction were permitted. Our presentation builds the interpretations
of ∧, ¬, ∨, → ∀ and ∃ into the rules governing truth-evaluation. The target is a notion of formal
consequence, where form is determined by which expression’s interpretations are held fixed. We also
need to insist on uniform interpretations, otherwise A wouldn’t imply A.
Supervaluations 69
true, despite mathematics plausibly not being part of logic.5 Following Stewart
Shapiro (1998), we therefore insert ‘necessarily’ between the ‘iff’ and quantifier
over interpretations.6 This avoids both kinds of problem, whilst remaining broadly
Tarskian in spirit. On this approach, Γ implies C iff the members of Γ and C cannot
be interpreted to make the former true and the latter false.7
The goal of model-theory is a mathematical representation of the space of pos-
sible interpretations. We want to define a class of mathematical structures S and a
relation R between members of S and sentences A such that, for any possible dis-
tribution π of truth-values across sentences, there is a structure x ∈ S such that:
R(x, A) iff π(A) = True. The structures represent interpretations. The condition
under which R(x, A) represents the truth-condition that A would have were the
interpretation(s) represented by x the actual interpretation.
Model-theory cannot be completely successful. Etchemendy’s example of the
Continuum Hypothesis provides one example why. Unrestricted quantification
provides another. In each model, the quantifiers range over a set. Since there is
no set of all sets, model-theory cannot capture quantification over all sets. Still,
these limitations shouldn’t matter for our purposes. Our primary interest is vague-
ness, not unrestricted quantification or the outer reaches of logical possibility.
In order to use the supervaluationist formalism for this purpose, two things are
required. First, we need to identify a class of elements of the formalism with the
class of possible interpretations.8 Second, we need to identify a dyadic relation R
with the true-under relation between sentences and possible interpretations I: A is
true under I iff A would be true if it had the content assigned it by I. We can then
identify the truth-condition of A under I with the defined condition under which
R holds between A and the representative of I.
5 The reason for this is that the Continuum Hypothesis may be formulated in a second-order
language that lacks non-logical vocabulary.6 In the relevant sense of necessity, it’s possible for a sentence to be interpreted thus-and-so iff the
language’s semantic/compositional structure doesn’t rule out it’s being interpreted thus-and-so.7 The key thesis of model-theoretic semantics is that there are enough mathematical structures
to represent every possible way of interpreting a language. Etchemendy’s criticisms show that this
assumption is false, and hence that there are representational limits on mathematised semantics.8 This is the sense of ‘identify’ in which identification is a form of representation: the items iden-
tified with interpretations are used to represent interpretations.
Supervaluations 70
Regarding the first task, there are two candidate accounts of interpretations.
(i) Supervaluationist models.
(ii) Sharpenings.
§2.3 and §2.4 consider these in turn, and the result of combining them with var-
ious candidate accounts of true-under. All bar two combinations of views will be
rejected. These are assessed in the remainder.
2.3 Interpretations as supervaluationist models
This section examines the identification of interpretations with supervaluationist
models. On this view, each vague language has a unique semantic structure, rep-
resented by a supervaluationist model. Vagueness is a feature of an expression’s
content; a feature of the propositions expressed by sentences featuring that expres-
sion. Thus clear truth (falsity) becomes a semantic feature of propositions, due to
the supervaluationist account of it as supertruth (superfalsity):
M makes A clearly true iff A is supertrue in M.
M makes A clearly false iff A is superfalse in M.
M makes A borderline iff A is neither supertrue nor superfalse in M.
We want to add an account of consequence to this. So we need to convert the
triadic s, M A relation into a dyadic M A to represent the true-under relation
between interpretations/supervaluationist models and wffs. There seem to be three
options:9
Particular M A iff a, M A, where a is a privileged sharpening in M.
Subtruth M A iff, for some sharpening s ∈ M : s, M A.
Supertruth M A iff A is supertrue in M.
Only the last of these stands up to scrutiny. We take them in turn.
9 Although there are other possibilities, this isn’t the place for an exhaustive survey. These are the
most obvious and popular candidates.
Supervaluations 71
2.3.1 Particular
The Particular account identifies truth with truth at a privileged sharpening in the
intended interpretation. This conflicts with our initial characterisation of super-
valuationism (p.64) by privileging one sharpening over all others when evaluating
for truth. It would be less misleading to identify interpretations with the privi-
leged sharpenings themselves, rather than supervaluationist models. (The result is
formally akin to Williamson’s (1994) epistemic view, described in §2.4.1.) So let us
set this option aside.
2.3.2 Subtruth
On this view, a sentence is true iff true under some way of making its constituents
precise. Dominic Hyde (1997) endorses this.
Consider a ball b which is a perfectly balanced red/orange borderline case. Both
the following are borderline, and hence neither supertrue nor superfalse:
b is red
b is orange
Since neither is superfalse, each is true at some sharpening. So both are true
simpliciter. But red and orange are incompatible. More generally, whenever A is
borderline/supertruth-valueless, both A and ¬A will be true. Yet that’s logically
impossible if ¬ expresses negation. The background picture is one on which the
semantic rules overdetermine truth-value in borderline cases, and the result is in-
consistency.
A response is available. Both ‘b is red’ and ‘b is orange’ can be supertruth-
valueless in a model without both being true at the same sharpening. So both can
be borderline without their conjunction ‘b is red ∧ b is orange’ being true at some
sharpening, and hence without being true. So it does not follow that their con-
junction will be true. The defender of Subtruth may therefore claim to respect the
incompatibility of red and orange by permitting no sharpening that places any-
thing in the extension of both ‘red’ and ‘orange’. (These penumbral connections
are discussed in §2.5.4.)
Supervaluations 72
Similarly, no sharpening makes both A and ¬A true. So even if these sentences
are both supertruth-valueless (and hence true), their conjunction A ∧ ¬A will be
superfalse and hence false. The defender of Subtruth may therefore claim to avoid
inconsistency by making no contradictions true.
These responses are unsatisfactory. They amount to observing that if truth is
subtruth, then Conjunction Introduction is unsound. One problem is that that
principle is too central to our understanding of conjunction to give up. Another
is that it doesn’t address the initial problems: b has incompatible colours; both A
and ¬A are true. The response blocks expression of these problems using conjunc-
tion. But we can still truthfully say “b is red. b is orange.” and pA. ¬Aq. These
are no less objectionable than the conjunctions in question; endorsing incompat-
ible claims using successive successive sentences is no better than doing so using
a single conjunctive sentence. We should therefore reject the Subtruth account of
truth.
2.3.3 Supertruth
This leaves the identification of truth with supertruth. Since a sentence is false
iff its negation is true, we also have the identification of falsity with superfalsity.
This is probably the most popular form of supervaluationism. Fine (1975), Hartry
Field (1974), Vann McGee and Brian McLaughlin (1994; 2000), and Rosanna Keefe
(2000) all endorse it in one form or another.
Since a sentence can be neither supertrue nor superfalse in M, by being true at
only some sharpenings in M, this view violates the classical semantic principle of:
Bivalence For any sentence A, either A is true, or A is false.
Since clear truth and truth are both identified with supertruth, it follows that all
borderline sentences lack truth-value. Each borderline sentence is a counterexam-
ple to Bivalence.
Plugging this view about truth into the Tarskian analysis of consequence gives
the relation Williamson (1994, p.148) calls global consequence:
Γ |=global C iff, for any model M, if every member of Γ is supertrue in M, then
C is supertrue in M.
Supervaluations 73
Consequence is preservation of truth-at-all-sharpenings in all models. Since sharp-
enings are classical models, all and only the classical logical truths are true at all
sharpenings. Hence all and only the classical logical truths are |=global-logical
truths. So even though Bivalence is false, the following classical logical law is
sound:
Law of Excluded Middle Every sentence pA ∨ ¬Aq is a theorem.
In fact, classical consequence and |=global coincide within predicate calculus. (Fine,
1975, p.125 sketches the proof.) Matters are more complex in languages enriched
with a clarity operator ∆, but we’ll come to that later (§2.6).
On the present view, each supervaluationist model represents an interpretation
of a vague language. Can we say anything more? What, for example, is the in-
terpretation of a predicate? Focus on monadic predicates for simplicity. What is
a monadic predicate’s contribution to truth-conditions, according to the present
view?
In classical semantics, each interpretation I assign a set of objects to each predi-
cate Φ, its I-extension. Φ’s I-anti-extension is the set of objects (in the domain) that
don’t belong to Φ. Atomic predications are interpreted using the following truth-
and falsity-conditions:
Φα is I-true iff the I-referent of α belongs to the I-extension of Φ.
Φα is I-false iff the I-referent of α belongs to the I-anti-extension of Φ.
Now, supervaluationist models are unlike classical interpretations because they
don’t actually assign any extra-linguistic semantic values to expressions: content is
determined by all the sharpenings in a model, rather than assigned to expressions
directly by supervaluationist models themselves. Nonetheless, we might seek to
generalise the classical picture as follows.
Let s, . . . , t be every sharpening in M. Define the M-extension of Φ as the inter-
section of JΦKs, . . . , JΦKt: the set of objects every sharpening assigns to Φ. Define
the M-anti-extension of Φ as the set of objects x ∈ Ds such that x /∈ JΦKs, . . . , x /∈
JΦKt: the set of objects no sharpening assigns to Φ. Now we can offer the following
account of atomic predications:
Supervaluations 74
Φα is M-true iff the M-referent of α belongs to the M-extension of Φ.10
Φα is M-false iff the M-referent of α belongs to the M-anti-extension of Φ.
Given the definitions of M-extension and M-anti-extension, Φα is M-true (M-false)
iff Φα is supertrue (superfalse) in M. This account therefore respects the identifi-
cation of truth (falsity) with supertruth (superfalsity). The view departs from the
classical picture because a predicate’s anti-extension is not determined logically,
solely on the basis of its extension. However, the classical conceptions of predica-
tion and of a predicate’s contribution to truth-conditions are retained.
This view is problematic. Sharpenings play no role in its truth-conditions for
atomic predications, yet they are indispensable to those for molecular statements.
The semantics of atomic and molecular statements is therefore non-uniform. Since
a uniform semantics is preferable, we should reject this account of the semantic
role of predicates and of the truth-conditions of atomic predications. Instead, a
predicate’s semantic contribution should be identified with its role in delimiting
the space of sharpenings as a whole. On this view, semantic relations between
expressions prevent the attribution to them of discrete semantic contributions.
The question now arises: what is a sharpening? The following five sections
address the following five answers in turn. Sharpenings are. . .
. . . ways a vague language could be made precise.
. . . ways a precise boundary could be drawn.
. . . classical interpretations.
. . . theoretical posits.
. . . artefacts of the formalism.
We will eventually settle on the last of these.
10 M-reference is unproblematic because we imposed the following constraint on models M: JαKs =
JαKt, for all sharpenings s, t ∈ M and singular terms α.
Supervaluations 75
2.3.3.1 Sharpenings as ways a vague language could be made precise
This is perhaps the most natural account of the supervaluationist formalism. It
seems to be endorsed by Keefe (2000, p.154), a prominent supervaluationist.11 The
idea is that a vague language can be made precise in many different ways. If every
way of making the language precise makes A true, then A is clearly true; if every
way makes A false, then A is clearly false; and if some ways make A true while
others make A false, then A is borderline.
This account of sharpenings faces two problems. The first is that it confuses
counterfactual semantic status with actual semantic status. This will not move
advocates of the view, however. For their core thesis is that actual truth is truth in
every possibility where the language is made precise.
The second problem arises from the fact that, “[t]o make an expression precise,
uncontroversial truths involving it must be preserved” (Keefe, 2000, p.154). This
is the key difference between making a meaning precise and replacing it with a
precise meaning. But as Jerry Fodor and Ernest LePore (1996) observe, sharpenings
do not respect all uncontroversial truths. Consider:
Everyone greater than 5’11” in height is tall
Each way of making ‘tall’ completely precise makes this true or makes it false. But
since it’s analytically borderline, and hence neither true nor false, no way of making
‘tall’ completely precise respects the meaning of ‘tall’. Consider also:
It’s borderline whether everyone greater than 5’11” in height is tall
Each way of making ‘tall’ completely precise makes this analytic truth false. So no
way of making ‘tall’ completely precise respects every uncontroversial (analytic)
truth involving it. So sharpenings are not ways a vague language could be made
completely precise.
Keefe replies that. . . :
“. . . this objection misrepresents the role of precisifications: such valua-
tions do indeed fail to capture all features of the meanings of our pred-
11 Keefe doesn’t carefully distinguish accounts of sharpenings. We’ll see in §2.3.3.3 that she en-
dorses different accounts on different pages.
Supervaluations 76
icates. . . . But this constitutes no objection to the theory, for the claim is
that it is quantification over all precisifications that captures the mean-
ing of the natural language predicates; the individual precisifications
need not.” (Keefe, 2000, p.190; original emphasis)
This response fails. The objection wasn’t to the identification of truth with su-
pertruth per se, but to combining that identification with the present account—
indeed, Keefe’s own account—of sharpenings. Simply reaffirming this combina-
tion of views does not make them consistent. So we should reject this account of
sharpenings.
2.3.3.2 Sharpenings as ways a precise boundary could be drawn
This option explains sharpenings in terms of boundary-drawing. Sometimes we
must decide whether to count a borderline case as a positive or negative case. Sup-
pose the owner of a record shop is re-organising her stock by genre. She encounters
a tricky case: should the John Adams records go in the minimalism section? A de-
cision is required one way or another, but competence with ‘minimalism’ enforces
neither choice. Suppose the store owner decides not to count Adams amongst the
minimalists. This has consequences for her classification of the remaining stock:
nothing less minimal than Adams counts as minimalism.12 There seems nothing
illegitimate about this commonplace aspect of linguistic practice.
The proposal is to treat sharpenings as formal representations of the effect of
such decisions on classification. Borderline cases are those that can be competently
called either way; that’s all there is to being a borderline case.
Three problems arise. Firstly, it is doubtful whether we can make classifica-
tory decisions that settle all borderline cases. Secondly, competence arguably man-
dates leaving a classificatory gap, however small (Shapiro, 2006, pp.8–12). Thirdly,
§2.7.2 argues that if borderline sentences lack truth-value, then borderline clas-
sificatory decisions are semantically illegitimate. If so, then sharpenings cannot
be explained in of legitimate such decisions. So we should reject this account of
sharpenings.
12 Similar issues arise for ‘less minimal than’.
Supervaluations 77
2.3.3.3 Sharpenings as classical interpretations
This view holds that a range of classical interpretations all fit the meaning-
determining facts equally well. None is privileged over any other as the actual,
or intended, interpretation of our language. Rather, these interpretations all con-
tribute jointly to the determination of (super)truth-conditions. Field (1974) and
Keefe (2000, pp.155–9) endorse views along these lines. Keefe also claims that it
fits the picture of vagueness as “semantic indecision” endorsed by Lewis (1986b,
p.212; 1993a).13 We will see that Keefe is wrong about this.
This view can explain the supervaluationist formalism in terms the classical
semanticist finds legitimate. It departs from classicality by first generalising the
classical metasemantic14 picture to accommodate meaning-determining facts too
coarse-grained to rule out all bar one classical interpretation. Semantic departure
from classicality then comes from identifying truth with truth on all intended clas-
sical interpretations.
The coherence of this view is doubtful. It involves two theses: (i) the meaning-
determining facts don’t determine a unique intended interpretation; and (ii) each
vague language possesses a unique intended interpretation. Taken at face value,
these are obviously inconsistent. The inconsistency is supposed to be resolved
by taking the interpretations in (i) as classical interpretations and those in (ii) as
supervaluationist models. Two problems arise. The first is that the metaseman-
tic concepts used to determine the classical interpretations enter into the truth-
conditional content of all ordinary vague expressions, if this view is correct. The
second is that determination of a unique intended vague interpretation (superval-
uationist model) should suffice to ensure that no precise (classical) interpretation
fits our use of language even approximately: if our use of language determines a
vague content, then no non-vague content is even a candidate content. The lesson
of this second point is that if truth is supertruth, then vagueness is not semantic
13 Note that Lewis does not combine the semantic indecision view the claim that truth is su-
pertruth. His (1970a, p.228) does propose that identification, but explicitly not in combination with
the semantic indecision picture.14 Metasemantics is the study of how expressions come to possess semantic properties, and what
those properties are.
Supervaluations 78
indecision.
On this account of sharpenings, supertruth is more naturally seen as a partly
semantic and partly metasemantic concept, rather than as the primary notion of
semantic evaluation, i.e. truth. §2.4 develops this kind of view. We set aside the
account of sharpenings as classical interpretations until then.
2.3.3.4 Sharpenings as theoretical posits
This view suggests that sharpenings should be treated like any other theoretical
posit, and ‘sharpening’ like any other theoretical term. We can make this precise
using a variant on Lewis’s (1970b) account of theoretical terms.
Let S be the supervaluationist theory (as formalised in predicate calculus). Let
S(X) be the result of replacing every occurrence of ‘is a sharpening’ in S with the
unbound predicate-variable ‘X’ (where ‘X’ does not occur in S). Then the property
of being a sharpening is whichever (unique) property satisfies S(X). The super-
valuationist theory S thereby provides an implicit definition of what it is to be a
sharpening. Sharpenings are like electrons in this respect: although ‘electron’ isn’t
explicitly definable using everyday vocabulary, electron-theory permits an implicit
definition of the electrons as whichever objects behave as it claims.
This leaves us with only descriptive knowledge of sharpenings, not partic-
ular knowledge: the sharpenings are whatever occupy the sharpening-role in
supervaluation-theory. Our knowledge of sharpenings is just like your knowledge
of quarpenings, if all you know about quarpenings is: quarpenings are what Nick
has in his pocket. This is no objection to the view. For sharpenings are no differ-
ent in this respect than any other theoretical posit. Furthermore, there are reasons
to think that a significant portion of our ordinary knowledge is also descriptive
(Lewis, 2009).
On this view, belief in sharpenings is justified to the extent that supervalua-
tionism as a whole is adequate. The better it can accommodate our use of vague
language, the better the justification for believing in sharpenings. The key diffi-
culty for the view is that we can identify truth with supertruth without incurring
ontological commitment to sharpenings or compromising supervaluationism’s ex-
Supervaluations 79
planatory ambitions. We examine this rival next.
2.3.3.5 Sharpenings as artefacts
This final view denies that sharpenings exist, other than as the pure mathemati-
cal subjects of the supervaluationist formal theory. McGee and McLaughlin (1994,
2000) and Josh Dever (2009) endorse views along these lines. Supervaluationism
is taken as a formal framework structurally similar to some aspects of vague lan-
guage, though not necessarily to all. Although some features of the framework
correspond to features of vague language, they need not all do so; and the present
view claims that sharpenings don’t.
Roy T. Cook (2002) and Shapiro (2006, ch.2), distinguish three attitudes to-
wards applied mathematical theories:
Representationalism Every feature of the formalism represents a feature of the
target system.
Modelling Some, but not necessarily all, features of the formalism represent fea-
tures of the target system.
Instrumentalism No features of the formalism represent features of the target sys-
tem (only input-output matching and predictive success matter).
Representationalism and Instrumentalism lie on extreme ends of a spectrum of
views. Both extremes are problematic.
Representationalism makes very high demands on the deployment of formal
tools. We want a mathematical representation of all possible assignments of log-
ically relevant content to a vague language. But there is no pre-theoretic reason
to expect that any mathematical structure will exactly resemble the structure of a
vague interpretation, especially since the structures in question have to be com-
prehensible to, and manipulable by, beings like ourselves. §2.9.6 even argues that
no mathematical structure will do so. So if Representationalism is correct, there is
no reason to expect that our project will be a success. Even the slightest departure
from perfection would render it a total failure.
Supervaluations 80
Instrumentalism is problematic because we would like more than mere empir-
ical adequacy and predictive success from our theorising. We would like to know
why a successful theory is successful. The most straightforward explanation is that
(at least some of) the theory’s internal structure corresponds to structure in the
target system. But that explanation is incompatible with Instrumentalism.
Modelling provides a moderate alternative to Representationalism and Instru-
mentalism, that avoids their worst excesses whilst accommodating their key in-
sights. Those aspects of the formalism that represent features of the target system
we call representors; those that don’t we call artefacts. Can we say anything general
and informative about which features of which theories fall into which category?
Well, our objection to Instrumentalism suggests that features essential to a theory’s
explanatory and predictive success shouldn’t be treated as artefacts. And the fol-
lowing version of Ockham’s Razor suggests that, ceteris paribus, only those features
should be treated as representors:
Posit as few kinds of entities as are necessary to explain a theory’s success.
It follows that exactly those features of the formalism necessary to explain a the-
ory’s success should be treated as representors. Are sharpenings amongst those
features of supervaluationism? Arguably not.
Sharpenings are needed to formulate tractable and comprehensible definitions
of the relations between truth-conditions which interpret the connectives. But they
need serve no other role. We can regard them as mere calculating devices used to
determine the truth-conditions of wffs, rather than as components of the semantic
structures of vague languages. McGee and McLaughlin (1994, §4) and Dever (2009,
§6) even offer accounts of the theoretical utility of sharpenings in terms of features
of the consequence relation for a vague language.
From the point of view of representational content, the resulting view treats a
supervaluationist model as a black box: information about the world goes in (in
the form of information about membership), and truth-values come out. Vague
interpretations either lack internal structure, or the supervaluationist formalism
does not capture it. Given the difficulties with the other accounts of sharpenings,
we will henceforth assume that if interpretations are supervaluationist models and
Supervaluations 81
truth is supertruth, then sharpenings are artefacts of the formalism.
2.3.4 Interpretations as models: concluding remarks
This section examined the identification of interpretations with supervaluationist
models. Views that don’t identify truth with supertruth were rejected in §§2.3.1–
2.3.2, and all bar one account of sharpenings were rejected in §§2.3.3.1–2.3.3.5.
This leaves only the view that combines these three theses:
(i) Interpretations are supervaluationist models.
(ii) Truth under an interpretation is supertruth in a supervaluationist model.
(iii) Individual sharpenings don’t represent anything in the semantic structures
underlying vague language.
The conjunction of (i)–(iii) we call the Supertruth View. Of all the views that endorse
(i), it is the only one we will consider in the remainder (although the following
discussion won’t turn on whether sharpenings are treated as artefacts or theoretical
posits). We now examine a different account of interpretations.
2.4 Interpretations as sharpenings
This section develops a view that identifies interpretations with sharpenings, rather
than with supervaluationist models. Vagueness is located in a language’s associa-
tion with a range of classical semantic structures (represented by the members of
a supervaluationist model). §§2.4.1–2.4.2 develop the metaphysics. §2.4.3 turns to
truth and consequence. §2.4.4 responds to an objection.
2.4.1 The association relation
The view is under-specified without an account of the association relation between
languages and classes of interpretations. What is it for a language to be associated
with many interpretations? What is the representational role of supervaluationist
models? By way of illustration, this section outlines Williamson’s (1994) epistemic
view.
Supervaluations 82
According to Williamson, each vague language possesses a unique intended
classical interpretation: vague sentences express unique classical propositions. But
there are many possibilities indiscriminable (to beings like ourselves) from actual-
ity and in which the intended interpretation of our language is not its actual inter-
pretation. For all we know, these possibilities could be actual; each of these coun-
terfactually intended interpretations could actually be intended. The borderline
cases are those over which these interpretations disagree. Vagueness thus becomes
a form of semantic ignorance.
Williamson’s epistemicism offers a philosophical account of the supervaluation-
ist formalism. Supervaluationist models represent epistemic states of typical lan-
guage users. The sharpenings within a model represent the interpretations that
typical speakers cannot know to be incorrect interpretations of their language; a
typical speaker’s true belief that one or other is the actual intended interpretation
could easily have been wrong. Clarity, as represented by supertruth, is thereby
analysed using both semantic and epistemic concepts: x is clearly F iff our imper-
fect semantic knowledge doesn’t prevent us from knowing that x is F. The rela-
tivisation of to models as well as sharpenings allows expression of claims whose
truth-value depends on the epistemic states of typical speakers as well as on the
interpretations of their expressions (though the expressive resources to do so must
await the clarity operator ∆ introduced in §2.5.1).
This account of the formalism conflicts with our initial characterisation of su-
pervaluationism by privileging one sharpening over all others in the determination
of truth-value. It does however, highlight the need for more detail about the repre-
sentational role of supervaluationist models before we have a fully-fledged theory
of vagueness. The following account of this missing detail draws inspiration from
Lewis’s brief and scattered writings on vagueness.
2.4.2 A Lewisian theory of association
In “General semantics”, Lewis writes:
“[W]e have so far been ignoring the vagueness of natural language. Per-
haps we are right to ignore it, or rather to deport it from semantics to the
Supervaluations 83
theory of language-use. We could say, as I do elsewhere [Lewis (1969,
ch.5)], that languages themselves are free of vagueness but that the lin-
guistic conventions of a population, or the linguistic habits of a person,
select not a point but a fuzzy region in the space of precise languages.”15
(p.228 Lewis, 1970a, my emphasis)
Lewis’s languages are our interpretations.16 Elsewhere he adds:
“Super-truth, with respect to a language interpreted in an imperfectly
decisive way, replaces truth simpliciter as the goal of a cooperative
speaker attempting to impart information. We can put it another way:
Whatever it is that we do to determine the ‘intended’ interpretation of
our language determines not one interpretation but a range of interpre-
tations. (The range depends on context, and is itself somewhat indeter-
minate.) What we try for, in imparting information, is truth under all
the intended interpretations.” (Lewis, 1993a, p.172)
The idea is to analyse vagueness via multiplicity of intended interpretation.
The meaning-determining facts settle which interpretations bear on linguistic
communication within a community of language-users. These are the intended
interpretations of that community’s language; they assign to its expressions the
content that members of the community express when using those expressions to
communicate with one another. We say that any such interpretation fits the com-
munity’s use of the language, or simply fits the community for short.
Think of the meaning-determining facts as determining a triadic relation: inter-
pretation x fits community y at least as well as interpretation z. This relation induces
an ordering on interpretations relative to a given community. An intended inter-
15 Lewis continues: “However, it might prove better to treat vagueness within semantics, and we
could do so as follows.” He then outlines a degree-theoretic version of the Supertruth View.16 Matters are a little more complex than this. Lewis’s languages are functions from sentences
onto truth-conditions. These languages are defined by grammars. A grammar G is a pair 〈SG, JKG〉,
where SG is a function from sentences onto syntactic structures and JKG is a function from the basic
constituents of these structures onto semantic values. Unlike Lewis, we assume a fixed syntactic
structure. Our interpretations are therefore the second elements of the class of grammars with the
appropriate shared first element.
Supervaluations 84
pretation is a greatest element in this ordering:17 no other interpretation fits the
community at least as well as it does. When the ordering is total, the community’s
language has a unique intended interpretation and their utterances express unique
propositional contents. When the ordering is only partial however, there may be
many greatest elements, each of which is an intended interpretation of the commu-
nity’s language.18 Utterances by members of such communities therefore express
multiple propositional contents.
This Lewisian approach to vagueness claims that vagueness is multiplicity of
intended interpretation. Our use of vague language is too coarse-grained to deter-
mine a total ordering on interpretations. The result is that utterances made using
our vague language expresses many propositional contents, so similar to one an-
other that ordinary usage doesn’t distinguish between them. A vague language has
not one but many semantic structures, each of which fits the meaning-determining
facts well enough to count as really giving the language’s content, and none of
which fits those facts better than any other.
We can now provide an account of the supervaluationist formalism. Models
represent states of the meaning-determining facts. Sharpenings represent inter-
pretations. A model M represents a sharpening s as a greatest element in the fit-
ordering (i.e. as an intended interpretation) iff s ∈ M. Different models represent
different ways the meaning-determining facts could bear on intended interpreta-
tions.
On this view, the semantics is classical. The departure from classicality is not
semantic but metasemantic: vague languages are multiply interpreted, they have
many intended interpretations instead of just one.
This provides an analysis of clarity. A sentence is clearly true (as used by a given
community) iff true under every intended interpretation (of the community’s lan-
guage); clearly false iff false under every intended interpretation; and borderline iff
17 Greatest elements must presumably also exceed some threshold in the fit-ordering in order to
count as intended: intended interpretations are those that fit well enough, as well as better than any
other.18 Think of a partial ordering as a branching tree-like structure. The intended interpretations are
the terminal nodes in the tree corresponding to the partial fit-ordering.
Supervaluations 85
true under some but not all intended interpretations. Clarity is thereby explained
using both semantic and metasemantic concepts.
2.4.3 Truth and consequence
We’ve got an account of the representational role of supervaluationist models, sharp-
enings and the relation between them. We want an account of consequence. So we
need an account of true-under. Models represent states of the metasemantic facts.
So we want to convert the triadic relation s, M A into a dyadic relation s A.
The extension of this relation will represent the extension of true-under, as it would
be were the metasemantic facts as M represents them. The options are analogues
of views considered earlier:
Particular Suppose the metasemantic facts are as M represents them. Then: s A
iff s, M A.
Subtruth Suppose the metasemantic facts are as M represents them. Then: s A
iff, for some sharpening t ∈ M, t, M A.
Supertruth Suppose the metasemantic facts are as M represents them. Then: s
A iff A is supertrue in M.
Plugging these into the Tarskian account of consequence result in different conse-
quence relations.
Subtruth succumbs to the logical objections in §2.3.2. Supertruth succumbs to
the objection in §2.3.3: it gives a non-uniform semantics for atomic and molecular
statements. Furthermore, any account of truth under an interpretation s that in-
volves reference to, or quantification over, interpretations other than s succumbs to
another objection: the metasemantic concepts that delimit the space of sharpenings
enter into the truth-conditional content of every vague sentence.
This leaves only the Particular account. Plugging this into the Tarskian analysis
of consequence gives what Williamson (1994, p.148) calls local consequence:
Γ |=local C iff, for any model M and sharpening s ∈ M, if s, M Γ, then
s, M C.
Supervaluations 86
‘s, M Γ’ means that s, M A, for all A ∈ Γ. |=local preserves truth under an
interpretation, whatever the state of the metasemantic facts. This account of |=local
is exactly analogous to the account of classical consequence in standard possible-
worlds semantics for modal logic. |=local is therefore classical. Classical logic and
semantics are preserved wholesale by this Lewisian approach.
2.4.4 An objection: monadic truth
This section considers an objection to this Lewisian account of vagueness. The view
associates each vague language with a class of intended classical interpretations. A
sentence is borderline iff true under some but not all of them. But since a sentence
is true iff true under its intended interpretation, and false iff false under its in-
tended interpretation, it follows that borderline sentences are both true and false,
which is impossible. So the relativised ‘A is true under s’ cannot be converted into
an un-relativised ‘A is true simpliciter’. The objection is that the Lewisian cannot
accommodate the monadic nature of truth.
As it stands, the objection is under-specified. It might concern either the monadic
English predicate ‘is true’, or the monadic property of being true . Either way, a re-
sponse is available. We take these disambiguations in turn.
2.4.4.1 The monadicity of ‘is true’
A response to the first disambiguation will provide an account of the extension of
the English ‘is true’. The most natural suggestion is:
Suppose the metasemantic facts are as M represents them. Then s, M pA
is trueq iff s, M A.
On this account ‘is true’ s-applies to exactly the sentences true under s.19 That
predicate therefore expresses the primary property of semantic evaluation of En-
glish sentences. Since those sentences express many contents, there are many such
properties (one for each correct of content expressed). It doesn’t follow that there
are many ways the world itself is because it’s entirely language- and sharpening-
19 We ignore Liar-like paradoxes for simplicity.
Supervaluations 87
independent which truth-conditions are satisfied. It’s just that since vague sen-
tences have many contents, there are many equally correct ways of evaluating them.
2.4.4.2 The monadicity of Truth
This second disambiguation of the objection claims that our Lewisian proposal is
incompatible with the monadicity of the property of Truth. We respond by dis-
tinguishing this property from the primary property by which sentences are evalu-
ated. The Lewisian proposal claims that this latter property is a relational property
defined in terms of the monadic property of Truth
Note first that sentences are just concrete objects, strings of sounds or marks.
These concrete objects aren’t intrinsically meaningful, and hence aren’t intrinsi-
cally suitable for semantic evaluation either. Sentences become suitable objects of
semantic evaluation only through their use by members of a linguistic community
to communicate with one another. This use bestows content upon sentences.
The moral is that the primary truth-bearers are not sentences themselves, but
the contents they express. The Lewisian account of vagueness makes no claims
about contents and their evaluation, but only about sentences and their evaluation.
It is therefore compatible with the monadicity of (propositional) Truth. Evaluating
sentences is a three-step process. The first step identifies their propositional con-
tent(s). The second step evaluates those propositions for Truth. The third step as-
signs truth-values to sentences, depending on whether they express true contents.
Sentential truth is thus a relational notion defined in terms of the non-relational
notion of propositional truth and the expression of propositions by sentences. The
Lewisian theory of vagueness simply claims that this expression relation may be
many-many: many sentences may express the same proposition, and a single sen-
tence may express many propositions. This prevents there from being a single
uniquely correct way of assigning truth-values to sentences. But since the Lewisian
view is compatible with their being a uniquely correct assignment of truth-values
to propositions, this is unproblematic.
Supervaluations 88
2.4.5 The story so far
Two accounts of vagueness are now in place. One is the Supertruth View from §2.3.
This associates each vague language with a unique vague intended interpretation—
viz. a supervaluationist model—at the cost of a non-classical semantics. The other
is the Lewisian proposal from §2.4. This maintains classical semantics by associ-
ating vague languages with many intended interpretations. We will call this the
Sharpening View. The rest of this chapter assesses their relative merits. The Sharp-
ening View will emerge as clearly preferable. The next two chapters apply it to
Unger’s and Lewis’s puzzles in two different ways.
2.5 Four benefits
Why should we believe either of our philosophical interpretations of the superval-
uationist formalism? This section presents four key benefits the supervaluationist
may claim: an analysis of clarity and borderline status (§2.5.1); a response to the
Sorites (§2.5.2); compatibility with classical logic (§2.5.3); respect for penumbral
connections between borderline cases (§2.5.4). Some of the problems for superval-
uationism examined in subsequent sections (§§2.6–2.9) show that not all of these
benefits are available on both the Supertruth View and Sharpening View; the Su-
pertruth View will fare particularly badly in this respect.
2.5.1 Analysing clarity
Supervaluationism offers both formal and philosophical analyses of clarity. We
present the formal analysis now, and the philosophical analyses below (§§2.5.1.1–
2.5.1.2).
Supervaluationism identifies clear truth with supertruth, clear falsity with su-
perfalsity, and borderline status with lack of supertruth-value. This allows intro-
duction of a clarity operator ∆ akin to the� of standard possible-worlds semantics
for modal logic:
v, s, M ∆A iff v, t, M A, for all sharpenings t ∈ M.
∆A formalises pIt is clearly the case that Aq. This semantics for discourse about
Supervaluations 89
clarity and borderline status is both familiar and tractable. Of course, the value of
this formal treatment is secondary to the philosophical insight it brings. But it is
an attractive feature nonetheless. So, what philosophical account of clarity can the
supervaluationist provide? That depends on whether the Supertruth or Sharpening
View is in question.
2.5.1.1 Clarity and the Supertruth View
The Supertruth View identifies truth with supertruth and falsity with superfalsity.
Given their account of clear truth as supertruth and clear falsity with superfal-
sity, it follows that truth is clear truth, falsity is clear falsity, and borderline sen-
tences fall down a truth-value gap. Borderline Fs fail to be F and fail to be not-F.
This amounts to an analysis of clarity using (relatively) familiar semantic concepts.
§2.7.1, §2.7.2 and §2.9.10 raise doubts about this analysis. This section raises a
different problem, followed by two attractive features of the account.
Borderline status is not the only potential source of truth-value gap. Other ex-
amples include reference-failure and incomplete stipulations. An adequate analy-
sis of clarity must distinguish vagueness-related truth-value gaps from those aris-
ing from other phenomena, if there are any. The natural account is that only
vagueness-related gaps result from cross-sharpening variation in truth-value. But
since our preferred account of sharpenings treats them as artefacts, that explana-
tion isn’t available. Treating sharpenings as theoretical posits doesn’t alleviate the
problem, but only makes the explanation circular: cross-sharpening variation in
truth-value is implicitly defined in terms of vagueness-related truth-value gaps.
The best the Supertruth theorist can achieve, it seems, is to list all other sources of
truth-value gap and claim: vagueness-related gaps result from none of these other
sources of gap. This isn’t entirely satisfactory. But let us simply bracket this worry
for the time being and move on.
This account of vagueness explains two aspects of our use of vague language.
The first is that, if A is borderline, then there’s something improper about the un-
qualified assertion “A”. The explanation is that in making an assertion (i) one
incurs commitment to that which was asserted, and (ii) one ought to communicate
Supervaluations 90
information about the state of the world. When A isn’t true, one who asserts “A”
therefore (a) commits themselves to something that isn’t true, and (b) violates a
norm governing assertion by communicating misleading information. §2.7.2 ar-
gues that this explanation is too strong.
The second feature of vague language explained by this semantic analysis of
clarity is the seeming misguidedness of investigating whether A when it’s known
that A is borderline. The explanation is that A’s borderline status precludes there
being a correct answer to the question of whether A by making neither A nor ¬A
true. Investigation into a known borderline claim A is misguided because that
knowledge is incompatible with knowing whether A. §2.8 argues that this expla-
nation fails.
2.5.1.2 Clarity and the Sharpenings View
The Sharpening View analyses clarity using a combination of semantic and metase-
mantic concepts: clear truth is truth under every intended interpretation; clear fal-
sity is falsity under every intended interpretation; and borderline status is variation
in truth-value across intended interpretations. Not all of a borderline sentence’s
propositional contents are true. This explains why we shouldn’t make unqualified
assertions using borderline sentences.
The previous section noted two related components to the practice of assertion:
(i) commitment to what one asserts; (ii) the communication of information. Each
component justifies the rule:
Assert only the truth.
Sentences that aren’t supertrue express false propositions. Those who make asser-
tions using borderline sentences therefore (a) commit themselves to falsehoods,20
and (b) communicate misleading information. Thus we have:
20 Why must using A to make an assertion bring commitment to all of the propositions A ex-
presses? Couldn’t it simply be indeterminate which of those many propositions the commitment was
to? The answer is that this blocks a non-circular analysis of vagueness via multiplicity of interpre-
tation. Given this requirement, commitment to every proposition expressed by the sentence used to
make the assertion follows from the idea that the meaning-determining facts don’t distinguish one
from amongst many (good enough) candidate intended interpretations.
Supervaluations 91
Use A to make an assertion only if A is supertrue.21
Unqualified assertions made using borderline sentences are illegitimate because
they violate this (derived) rule.22,23
This explains why “Yes” and “No” are equally inappropriate answers to the
question “A?”, when A is borderline. For answering “Yes” amounts to the assertion
“A”. And answering “No” amounts to the assertion “¬A”. Since neither A nor
¬A is supertrue, neither is assertable. So neither “Yes” nor “No” is an appropriate
answer to the initial question. We can use this to explain borderline ignorance and
the misguidedness of investigation into known borderline claims.
There are two candidate necessary conditions on the s-truth of pS knows that
Aq:
The proposition s assigns to A is true.
Any proposition that any intended interpretation s′ assigns to A is true.
The first ensures that when A is borderline, pS knows that Aq is no better than bor-
derline. The second ensures that when A is borderline, pS knows that Aq is clearly
false. Either way, pS knows that Aq is not supertrue, and hence not assertable when
A is borderline. A’s borderline status excludes knowledge whether A in the sense
of making it in-principle illegitimate to claim to know that A and also illegitimate
to know that ¬A. Known borderline status makes investigation misguided because
that knowledge is incompatible with legitimately claiming to know the result of
the investigation.
21 Sentences can be true under varying proportions of intended interpretations. So one violation
of this rule can be worse than another. Thus one borderline assertion can be in worse standing than
another.22 Could the extension of ‘asserted’ vary across sharpenings in the following manner?: if S used
A to make an assertion, then ‘S asserted that p’ is s-true iff A s-expresses that p. This makes it
borderline whether, rather than clearly false that, one obeys the first rule above when one makes an
assertion using a borderline sentence. But there remain intended interpretations on which one fails
to adhere to the rule. So using borderline sentences to make assertions complies with the norms
governing assertion less well than does using supertrue sentences to do so, though it complies better
than does using clearly false sentences. This may explain why borderline assertions are better than
clearly false ones.23 Dorr (2003, §1 and pp.104–5) presents a related argument for a similar conclusion.
Supervaluations 92
2.5.2 The Sorites
The challenge of the Sorites is to explain why the seemingly valid argument from
the following three apparent truths to contradiction is unsound, and also why it
appears to be sound:
R1 Ten seconds after his birth, Russell was young.
R2 Forty years after his birth, Russell was not young.
R3 If Russell was young i seconds after his birth, then he was also young i + 1
seconds after his birth.
R1 and R2 really are indubitable: unless ‘young’ is trivial, some premisses of these
forms are true. The argument from R1–R3 to contradiction is classically valid
and formalisable in predicate calculus, hence both |=global- and |=local-valid. The
fault lies in R3: every complete sharpening makes some instantiation of R3 false.24
Hence R3 is clearly false. So why does it seem compelling?
Note first that since it’s vague when Russell ceased to be young, the location of
the young/not-young distinction varies across sharpenings. Let ‘young(x, y)’ for-
malise ‘x was young y seconds after x’s birth. Then no instantiation of the following
is supertrue:
(1) ∃i(young(Russell, i) ∧ ¬young(Russell, i + 1))
The Supertruth theorist may now offer the following conjecture:
Typical speakers reason as if a true existential generalisation required a true
instantiation.
This can’t be quite right because not everything is named.25 But we’ll assume that
everything is named because it simplifies exposition without affecting anything of
substance. The Supertruth theorist’s conjecture amounts to the claim that typical
speakers reason classically with existentials. This has the attractive feature of treat-
ing them as reasoning in accord with a natural and simple (i.e. classical) semantics.
24 We assume for simplicity that everything has a (clear) name. Nothing of substance turns on this.25 Our introduction to the Sorites in §1.2.1.1 contains discussion related to the inadequacies of this
conjecture.
Supervaluations 93
Assume that ordinary speakers detect that (1) lacks a true instantiation. The
Supertruth theorist’s conjecture then implies that typical speakers will regard (1)
as false. Since the negation of (1) is equivalent to R3, the Supertruth theorist has
an account of why R3 seems compelling: typical speakers infer it by an invalid (but
natural) argument from the correct observation that no instantiation of (1) is true.
The Sharpening theorist can’t quite adopt this account because they reject the
identification of truth with supertruth: every true existential does have a true in-
stantiation. They could offer an alternative conjecture:
Typical speakers reason as if a true existential generalisation required a clearly
true instantiation.
Since (1) lacks a clearly true instantiation, this will achieve the same result. But this
alternative conjecture cannot be justified by attributing classical inferential prac-
tice to ordinary speakers. In effect, the conjecture treats speakers as mistaken about
the content of their existentially quantified claims, but fails to explain why the mis-
take arises. Luckily for the Sharpening theorist, two alternative explanations are
available.
The first explanation begins by observing that (1) lacks a clearly true instan-
tiation. The previous section offered an account of why borderline status makes
(unqualified) assertion illegitimate. So no instantiation of (1) is assertable. Given
an argument from unassertability to falsity, the Sharpening theorist may then ex-
plain the apparent falsity of (1), and hence truth of R3, by attributing reasoning in
accord with that argument to typical speakers. One such argument is as follows.
The unassertability of instantiations of (1) isn’t the result of our own limita-
tions. No amount of investigation into Russell’s gradual aging could make any
such instantiation assertable. But when unassertability isn’t the result of our own
limitations, then that’s because there’s no truth there to assert.26 So every instan-
tiation of (1) must be false. So (1) must be false. So R3 must be true. Attributing
this line of argument to ordinary speakers treats them as ignorant (or forgetful) of
26 A counterexample: no instantiation of ‘some mammal was the first unnamed dog born at sea’ is
assertable (or even true), and not because of our own limitations; yet there’s no temptation to regard
it as false. However, since this contains semantic vocabulary within the scope of a quantifier, the
Sharpening theorist may claim that it is relevantly disanalogous to (1).
Supervaluations 94
another source of unassertability: the expression of multiple contents, only some
of which are true.
The second explanation for the attraction of R3 begins by observing that Russell
ages gradually. No single-second duration in his aging is more intrinsically signifi-
cant than any other. Neither, the Sharpening theorist claims, do successive single-
second durations stand in significantly different relations to our use of ‘young’.
R3 is a natural, though incorrect, way of reporting the following consequence of
those two facts: no single-second duration marks any significance difference as to
whether ‘is young’ applies to Russell, as that predicate is used within the commu-
nity of speakers of English.
2.5.3 Classical logic
§2.3.3 and §2.4.3 noted that |=local and |=global coincide with classical consequence
within predicate calculus. Hence both the Supertruth View and the Sharpening
View preserve classical logic when reasoning within languages of that form. This
doesn’t however, extend to languages containing ∆. More on that shortly (§2.6).
Now, classical logic isn’t absolutely mandatory. But it is (a) the default, (b) re-
quired for large portions of best science and mathematics, and (c) hard to see how
it could fail. Any departure from classical logic had therefore better (i) have strong
positive arguments in its favour, (ii) validate classical reasoning in mathematics,
and (iii) provide insight into the underlying semantic features of the language re-
sponsible for non-classicality. Since supervaluationism doesn’t satisfy (i) or (iii), it’s
a good thing that it preserves classical logic.27
2.5.4 Penumbral connection
Let b be a ball that’s a perfectly balanced red/orange borderline case. It would
be misleading to describe b without qualification as red, or to describe it without
qualification as orange. Nonetheless, it wouldn’t be misleading to describe it with-
27 Why doesn’t supervaluationism satisfy (i)? Because the argument in its favour is that it provides
an account of vagueness that can accommodate the data. The apparent validity of classical reasoning
is part of the data. What’s needed for (i) is a direct argument for non-classical logic from an account
of the nature of vagueness. The supervaluationist hasn’t provided this.
Supervaluations 95
out qualification as either red or orange. For how could b fail to be red or orange?
It’s clearly coloured. And it’s clearly no colour that’s neither red nor orange. So
surely it’s either red or orange, despite being neither clearly red nor clearly orange.
This is an instance of a penumbral connection: an analytic connection between the
borderline regions of vague predicates.28
Supervaluationists accommodate penumbral connections by imposing penum-
bral constraints: conditions on the extensions sharpenings assign to penumbrally
connected expressions. The intended interpretation(s) of a vague language respects
these constraints. For example, the following constraint makes ‘b is red ∨ b is or-
ange’ clearly true, even if both disjuncts are borderline:
For any sharpening s, either b ∈ JredKs or b ∈ JorangeKs, but not both.
Now consider a cube c that’s just slightly redder than b, though still borderline
red/orange. This should be clearly false:
c is orange ∧ b is red
For how could something redder than b fail to be red, if b is? And this should be
borderline:
c is red ∧ b is red
Yet the conjuncts of both are borderline. The following gives the desired result:
For any objects x, y and sharpening s, if x is redder than y, then: if y ∈ JredKs,
then x ∈ JredKs.
This second case also shows that the clear-truth-value of a molecular sentence
shouldn’t be a function of those of its components. For the conjunctions above
differ in clear-truth-value, though their conjuncts don’t. Since supervaluationist
logic isn’t clear-truth-value-functional, this counts in its favour. In fact, the failure
of clear-truth-value-functionality arises for any view that preserves classical logic
alongside borderline sentences. For when A is borderline, so are the following: A;
A ∨ A; A ∧ A; and ¬A. But A ∨ ¬A will then be clearly true and A ∧ ¬A clearly
28 This section draws on the argument against clear-truth-value-functionality in Edgington (1997,
§3).
Supervaluations 96
false.
We now turn from supervaluationism’s benefits to its problems.
2.6 A logical problem
Retention of classical logic was presented as one of supervaluationism’s key bene-
fits (§2.5.3). Williamson (1994, §5.3) challenges this. Although his argument isn’t
quite conclusive, we’ll see that it does force surprising revisionary theses on the
Supertruth theorist that many will find unacceptable. The Sharpening View, by
contrast, survives unscathed.
§2.6.1 presents Williamson’s objection. McGee and McLaughlin’s response is
examined in §2.6.2, alongside two difficulties for it in §§2.6.3–2.6.4. §2.6.5 closes
by showing how the Sharpening View evades the objection.
2.6.1 The argument
This section presents the argument against the classicality of supervaluationist
logic.
In what sense does the Supertruth View preserve classical logic? Since truth is
identified with supertruth, consequence is identified with |=global:
Γ, A |=global C iff, for any supervaluationist model M, if every member of
Γ ∪ {A} is supertrue in M, then C is supertrue in M.
Let |=cl be classical consequence. Say that an argument is |=x-valid iff its conclusion
is an |=x-consequence of the set of its premisses. Say that an inferential pattern is
|=x-sound iff it licenses only |=x-valid arguments. |=global and |=cl coincide within
predicate calculus. All |=cl-valid arguments formalisable within predicate calculus
are therefore |=global-valid. Hence any inferential pattern within predicate calculus
that is |=cl-sound is also |=global-sound. This is the sense in which the Supertruth
View preserves classical logic.
This fails for languages enriched with ∆. Williamson gives counterexamples to
contraposition, argument by cases, reductio and conditional proof. We’ll focus on the
Supervaluations 97
last of these:
Conditional Proof (CP) If Γ, A ` C, then: Γ ` A→ C.
‘Γ, A ` C’ means that C is derivable from the wffs in Γ together with A. Given
this rule of proof, the conditional A → C is derivable from premisses Γ whenever
C is derivable from Γ together with A. CP encodes the primary means of drawing
conditional conclusions from categorical premisses. CP is |=x-sound iff:
If Γ, A |=x C, then: Γ |=x A→ C.
CP is |=global-unsound because:
A |=global ∆A, but: 6|=global A→ ∆A.29
Although A |=global ∆A, the conditional A → ∆A is not a |=global-logical truth.
Here’s the proof.
First, we show A |=global ∆A. Suppose A is supertrue in M. Then s, M A, for
all s ∈ M. So by the rule for ∆: s, M ∆A, for all s ∈ M. So ∆A is supertrue in
M. So if A is supertrue in M, then ∆A is supertrue in M. Since M was arbitrary:
A |=global ∆A. This result shows that the following is |=global-sound:
∆In A ` ∆A.
We now show 6|=global A → ∆A. Suppose that (i) A is not supertrue in M, and
(ii) A is not superfalse in M. From (i): s, M 1 A, for some s ∈ M. By the rule for ∆:
s, M 1 ∆A, for all s ∈ M. But from (ii): s, M A, for some s ∈ M. Putting these
together: s, M A and s, M 1 ∆A, for some s ∈ M; let s∗ be such a sharpening.
By the rule for→: s∗, M 1 A → ∆A. So: A → ∆A is not supertrue in M. Hence:
6|=global A→ ∆A.
The problem comes from combining ∆In with CP. Suppose A. By ∆In: ∆A. By
CP: A → ∆A. Since A was our only premiss: ` A → ∆A. But since 6|=global A →
∆A: the unpremissed argument for A → ∆A is |=global-unsound. Any deductive
system containing both ∆In and CP is therefore |=global-unsound. We could avoid
29 J.R.G. Williams (2008) argues that this result is an artefact of an impoverished formal setting
that does not hold in a more satisfactory framework. There isn’t space to discuss Williams’s view
here. For a response to Williams, see my (forthcoming).
Supervaluations 98
this by excluding ∆In. But since that rule is sound, this serves only to artificially
block derivation of some consequences of our premisses. The Supertruth theorist
should therefore deny that CP is sound.
2.6.2 Restricting CP
McGee and McLaughlin (1998, 2004) object. They grant that ∆In is |=global-sound,
but deny that CP as formulated is part of classical logic. They claim that a restricted
version of CP that disallows ∆In within the scope of suppositions is all that classical
logic requires. Write ‘Γ, A `MM C’ when C is derivable without using ∆In from the
members of Γ together with A. McGee and McLaughlin claim that classical logic
requires not CP, but:
Restricted Conditional Proof (RCP) If Γ, A `MM C, then: Γ ` A→ C.
The idea is that although |=cl and |=global come apart, this doesn’t mandate revisions
to classical inferential practice because that practice requires only RCP, not CP. We
now have to ask: what is classical logic?
According to Williamson (2004, p.120), classical logic comprises “those forms
of logical inference tried and tested in mainstream mathematics and other branches
of science.” McGee and McLaughlin (2004, p.133) agree. This brings out two
virtues of retaining classical logic. The first is that if a classical rule is unsound,
then those parts of science and mathematics that employ it become suspect. The
second is that an inference rule’s successful employment throughout our best sci-
ence and mathematics provides inductive grounds for believing it successful else-
where in science, including semantics.
These virtues belong to any semantic theory that renders sound those inferences
required by best science and mathematics. McGee and McLaughlin claim that stan-
dard predicate calculus without ∆ suffices for formalising those inferences. Since
|=global and |=cl coincide within predicate calculus, the Supertruth View possesses
the virtues attendant upon retaining classical logic. Although those relations di-
verge in languages enriched by ∆, this doesn’t bring objectionable revisions to clas-
sical inferential practice.
Supervaluations 99
This dispute about the extent of classical logic concerns which of the following
dispositions is manifested in mainstream scientific and mathematical reasoning:
The disposition to conclude A → C on the basis of any derivation of C from
A.
The disposition to conclude A → C on the basis of any derivation of C from
A that doesn’t employ ∆In.
McGee and McLaughlin must claim that only the latter, weaker disposition is man-
ifested in the inferential behaviour of mainstream scientists, if they are to defend
RCP over CP. But it is not clear which of these competing accounts is preferable.
Attribution of the second disposition is the minimum required to explain the data,
if McGee and McLaughlin are right that mainstream science requires only those
inferences formalisable in predicate calculus. But then attribution of the first dis-
position avoids attributing to scientists restrictions on when they are prepared to
draw conditional conclusions, when those restrictions aren’t manifested in their
actual reasoning. It is therefore unclear whether or not the attempted restriction
of classical logic to RCP is successful.
2.6.3 The justification for RCP
RCP is prima façie objectionable. Why is ∆In inapplicable to mere suppositions?
The rule is sound: the language can’t be interpreted so as to make A true without
also making ∆A true. Thus ∆A may be inferred from the believed premiss A.30 So
if ∆A cannot be inferred from the mere supposition A, then that must be because
the content of A differs between premisses and suppositions: supposing A is not
the same as supposing A to be true. This creates two problems. Firstly, an account
is required of the content of supposing A, if it isn’t the same as supposing A to be
true or as believing A. Secondly, it undermines the role of deduction from suppo-
sitions in justifying belief in the conclusion of an argument on the basis of (i) belief
in its premisses and (ii) a prior deduction of that conclusion from the supposition
30 Unless believing A and believing A to be true have the same content, then it’s obscure what
role validity—i.e. truth-preservation under every interpretation—has to play in constraining rational
belief-formation.
Supervaluations 100
of the premisses. This section finesses McGee and McLaughlin’s objection to CP in
order to avoid these complaints.
Consider a thinker who supposes A and applies ∆In to derive ∆A. Granted that
her derivation was sound, what exactly has she shown? Since ∆In is |=global-sound,
this derivation shows that any model where A is (super)true is a model where ∆A
is also (super)true. It follows that A → ∆A is (super)true in all such models. Does
it follow that this conditional is (super)true in all models? That’s what’s needed for
|=global A → ∆A, and hence for our thinker’s derivation to license the conclusion
A→ ∆A outside the scope of the initial supposition.
This would follow if the only models where A isn’t (super)true were models
where A is (super)false. For A → ∆A is vacuously (super)true in any such model.
But these aren’t the only other models: some models make A neither (super)true
nor (super)false. Ensuring that A → ∆A is (super)true in all models where A is
(super)true and also (super)true in all models where A is (super)false, therefore
doesn’t suffice to ensure that A is (super)true in all models. Yet that’s all that’s
ensured by the |=global-validity of our thinker’s derivation of ∆A from A. CP fails
because in fact, if A is neither supertrue nor superfalse in M, then (i) s, M 1 ∆A,
for all s ∈ M, and (ii) s, M A, for some s ∈ M. From (i) and (ii) it follows that
s, M 1 A → ∆A, for some s ∈ M, and hence that A → ∆A is not supertrue in M.
This makes M a countermodel to |=global A→ ∆A. Hence McGee and McLaughlin’s
claim that assuming unrestricted CP is tantamount to assuming Bivalence (McGee
and McLaughlin, 2004, pp.134–5).
This finesses the restriction on CP. The use of ∆In within the scope of supposi-
tions is unproblematic. Likewise for drawing conditional conclusions on the basis
of such uses of ∆In. The problem lies in discharging the premiss/supposition to
which ∆In was applied, to yield a conditional without dependence on that pre-
miss/supposition. The |=global-soundness of ∆In ensures the truth of A→ ∆A only
under the supposition of A or the supposition of ¬A, not under any supposition
whatsoever; specifically, not under the supposition ¬∆A ∧ ¬∆¬A. By discharging
the supposition, we lose any record of this information. Hence we cannot do so.
No difference in A’s content when taken as a premiss or supposition is required
because there’s no difference in the applicability of ∆In to premisses and supposi-
Supervaluations 101
tions. The objection to RCP with which this section began therefore fails.
This suggests a sense in which supervaluationist semantics preserves classi-
cal reasoning, even within languages enriched by ∆. The countermodels to the
classical logical laws can be ignored when reasoning under the (possibly tacit) as-
sumption of precision—i.e. the assumption that there are no borderline cases and
Bivalence holds—because those countermodels are all models that make A value-
less. Provided we can exclude circumstances in which A is borderline from the
circumstances our reasoning must take account of, we can reason classically in a
language with supervaluationist semantics. It is certainly plausible that we can
exclude borderline cases within pure mathematics. And the replacement of vocab-
ulary susceptible to borderline cases with new classifications is arguably also one
of the hallmarks of science. Those, such as mainstream scientists and mathemati-
cians, who reason under the assumption of precision or in circumstances in which
borderline cases cannot arise, may therefore reason entirely classically.
2.6.4 A problem for RCP
This section argues that replacing CP with RCP brings unexpected and revisionary
consequences.
Given the following pair, RCP’s restriction to derivations that don’t employ ∆In
is no restriction at all:
(i) Logically valid deductions license the truth of English conditionals, whether
subjunctive or indicative.
(ii) English conditionals imply their corresponding material conditionals.
On the Supertruth View, A logically implies ∆A. So by (i): if it were that A, it
would be that ∆A. Then by (ii): A → ∆A. So any valid argument from A to ∆A
also licenses a valid argument for A→ ∆A that doesn’t employ ∆In. Hence RCP is
equivalent to CP. Those who reject CP but not RCP must therefore reject (i) or (ii). I
don’t know which is preferable, but neither is attractive and both assumptions are
commonplace. Following are three examples.
First example: Ian McFetridge (1990) assumes (i) without argument when ar-
guing that logical necessity is the strongest form of necessity.
Supervaluations 102
Second example: the portion of (i) that concerns counterfactuals follows from
the Lewis-Stalnaker semantics. Since the logical consequences of A are true at
any world where A is true, those consequences are also true at the closest world(s)
where A is true.
Third example: Williamson (2007, pp.293–4, 300) calls the portion of (ii) that
concerns counterfactuals “immensely plausible”, noting that it is an axiom of Lewis’s
logic for counterfactuals (Lewis, 1986a, p.132).
These appeals to the authority of classical logicians don’t show that (i) and (ii)
are true. But they do expose the Supertruth View’s (well hidden) revisionary im-
plications.
2.6.5 The Sharpening View
The Sharpening view renders CP sound. On that view, consequence is local conse-
quence:
Γ, A |=local C iff, for any model M and sharpening s ∈ M, if s, M Γ ∪ {A},
then s, M C.
∆In is |=local-unsound. For suppose that (i) A is not supertrue in M, and (ii) A is
not superfalse in M. By (i) and the rule for ∆: s, M 1 ∆A, for any s ∈ M. By (ii):
s, M A, for some s ∈ M. Putting these together: s, M A and s, M 1 ∆A, for
some s ∈ M. Hence: A 6|=local ∆A.
In fact, the Sharpening View’s formal treatment of truth, consequence and ∆ is
exactly analogous to that of truth, consequence and � in standard possible-worlds
semantics for modal logic. On the Sharpening View, vagueness mandates no more
nor less deviation from classical logic than does modality.
2.6.6 Supervaluationist logic: concluding remarks
We’ve seen two kinds of problems for the Supertruth theorist’s claim to preserve
classical logic. The first is that it’s unclear whether mainstream scientific reasoning
manifests disposition to reason in accord with the restricted or unrestricted ver-
sion of CP (§2.6.2). This first problem is somewhat ameliorated by the fact that
Supervaluations 103
unrestricted classical reasoning is permissible when, as in most science and math-
ematics, the possibility of borderline cases may be discounted (§2.6.3). The second
is that it brings revisionary consequences for the interaction of logical consequence
with English and material conditionals (§2.6.4). Although neither problem is deci-
sive, they are costs of the Supertruth View that aren’t incurred by the Sharpening
View (§2.6.5).
2.7 Two semantic problems
Let us turn from supervaluationist logic and inference to the semantics on which
they are based. §2.7.1 examines the Supertruth theorist’s notion of truth. §2.7.2
presents a difficulty in accommodating our apparent discretion about borderline
classification. The Supertruth View will again be shown to fare significantly less
well than the Sharpening View.
2.7.1 Truth and supertruth
Because borderline sentences are neither (super)true nor (super)false, the Supertruth
View violates:
Bivalence For any sentence A, either A is true or A is false.
But since sharpenings are just classical models, every classical theorem is true at
each sharpening and hence supertrue in each model. The following is therefore
sound:
Law of Excluded Middle (LEM) Every instance of pA ∨ ¬Aq is a theorem.
LEM is one component of the Supertruth theorist’s claim to preserve classical logic.
This section shows that this combination of views makes the identification of truth
with supertruth problematic.
2.7.1.1 Supertruth and Bivalence
The Supertruth View implies that A and pA is trueq differ in content. pA is trueq
should be true at a sharpening iff A is supertrue:
Supervaluations 104
s, M pA is trueq iff A is supertrue in M; iff t, M A, for all t ∈ M.31,32
Suppose A is neither supertrue nor superfalse in M. So s, M 1 A, for some s ∈ M.
By the rule for ‘is true’: s, M 1 pA is trueq, for all s ∈ M. So pA is trueq is
(super)false in M. But since A is not (super)false in M but valueless, A and pA is
trueq differ in (super)truth-value and therefore also differ in content.
Recall the Supertruth View’s account of clear truth as truth and clear falsity
as falsity (§2.5.1.1). This account of clarity is incomplete without an account of
the difference in content between A and pA is trueq. The Supertruth View’s own
account of clarity is the source of this extra explanatory burden. For that account
combines with the possibility of borderline cases to undermine Bivalence; and non-
Bivalence is responsible for the difference in truth-status of A and pA is trueqwhen
A is borderline. Without an account of the difference in content between A and pA
is trueq, the Supertruth View’s explanation of clarity in semantic terms is only a
pseudo-explanation.
Let us be clear about just what this shows. It does not show that the truth of pA
is trueq requires anything more than the truth of A, or vice versa; for the models
that make A true are exactly those that make pA is trueq true. The difference in
content comes from models that make A valueless and pA is trueq false. Although
the truth of pA is trueq requires no more nor less than the truth of A, its falsity
requires less than the falsity of A. This is what needs explaining.
2.7.1.2 From LEM to Bivalence
What account of this difference in content between A and pA is trueqmight the Su-
pertruth theorist offer? This section examines their conception of sentence-content,
arguing that it is unclear how such an account might proceed.
Classical semantics conceives sentence-content as comprising a truth-condition.
31 The alternative is: s, M pA is trueq iff s, M A. But then each instance of Bivalence is true
at each sharpening: A is true ∨¬(A is true). This either (i) reinstates Bivalence, or (ii) prevents the
object-language ‘is true’ from expressing genuine (super)truth.32 We use a truth-predicate and ignore the semantic paradoxes for simplicity. We could could just
as easily use a truth-operator and forego reference to sentences, in which case the paradoxes couldn’t
arise.
Supervaluations 105
If the condition is met, the sentence is true, otherwise it is false. The Supertruth
View cannot accept this. For if truth-conditions exhaust sentence-content, then
there is no space to distinguish the untruths that are false from those that fall down
a truth-value gap: in both cases, the condition that exhausts the content of the sen-
tence in question is unsatisfied. An alternative conception of sentence-content is
required.33
The Supertruth theorist needs to conceive sentence-content as comprising two
independently settled components: truth-conditions and falsity-conditions. A sen-
tence is true iff its truth-condition is met, false iff its falsity-condition is met, and
neither true nor false when neither condition is met. When the content-determining
facts co-operate, truth-conditions and falsity-conditions will partition the possibil-
ities and the (interpreted) sentence in question will be Bivalent. But the meaning-
determining facts need not co-operate; in which case, truth- and falsity-conditions
won’t partition the possibilities and the sentence won’t be Bivalent; in some pos-
sibilities, it will be neither true nor false because neither its truth-conditions nor
falsity-conditions are satisfied.
We can adapt an argument of Williamson’s (1997, §1) to make trouble for this
view. Since truth just is the satisfaction of truth-conditions, and falsity just is the
satisfaction of falsity-conditions, the following are analytic:
TC A’s truth-condition is satisfied iff A is true.
FC A’s falsity-condition is satisfied iff A is false.
Take ‘grass is green’ as an example. A plausible truth-condition is that grass is
green. And a plausible falsity-condition is that grass is not green. By LEM: grass
is green ∨ grass is not green. Suppose that grass is green. By TC: ‘grass is green’
is true. By ∨-introduction: ‘grass is green’ is true ∨ ‘grass is green’ is false. Now
suppose that grass is not green. By FC: ‘grass is green’ is false. By ∨-introduction:
33 What about (putative) truth-value gaps resulting from non-referring singular terms? The route
from reference-failure to valuelessness conceives of singular terms as contributing their referent to
truth-conditions. When no referent is contributed, no truth-condition is determined. Hence sen-
tences containing non-referring terms lack truth-conditional content. The Supertruth theorist, by
contrast, conceives gappy vague sentences as having a gappy content.
Supervaluations 106
‘grass is green’ is true ∨ ‘grass is green’ is false. So either way: ‘grass is green’ is true
∨ ‘grass is green’ is false. Since ‘grass is green’ was arbitrary, we can generalise to
Bivalence. Since TC and FC are analytic and we made no non-analytic assumptions:
Bivalence is analytic.
The Supertruth theorist must resist. How? Not by attacking the reasoning.
That requires only argument by cases, universal generalisation, modus ponens and
∨-introduction. The last three of these are |=global-valid. Argument by cases is trick-
ier. Although |=global-valid within predicate calculus, the following result shows
that it fails in languages containing ∆:
|=global A ∨ ¬A.
A |=global ∆A.
¬A |=global ∆¬A.
6|=global ∆A ∨ ∆¬A.
On the Supertruth View, ∆ amounts to an object-language reflection of (super)truth.
This result might therefore appear to cast doubt on arguing by cases from TC, FC
and LEM to Bivalence. This appearance is misleading.
A |=global ∆A and ¬A |=global ∆¬A together ensure that one of ∆A, ∆¬A will
be supertrue in any model where one of A, ¬A is supertrue. Hence ∆A∨ ∆¬A will
also be supertrue in any such model. But this is silent about models where neither
A nor ¬A is supertrue. Since A ∨ ¬A is supertrue in some models where neither
disjunct is supertrue, A |=global ∆A and ¬A |=global ∆¬A do not by themselves
ensure that ∆A∨∆¬A is supertrue in all models where A∨¬A is supertrue. Hence
even given |=global A ∨ ¬A, they do not ensure that ∆A ∨ ∆¬A is supertrue in all
models whatsoever. The countermodels to the validity of arguing by cases arise
because both ∆A and ∆¬A are in fact false at all sharpenings in all models where
neither A nor ¬A is supertrue. Nothing similar applies to the argument by cases
from TC and FC to Bivalence.
On the Supertruth View’s bipartite conception of sentence-content, the contents
of pA is trueq and of pA is falseq are exhausted by A’s truth-conditions and falsity-
conditions respectively. Thus TC and FC should be understood as analytically true
Supervaluations 107
material biconditionals, not mutual entailments.34 Since TC and FC are analytic,
they should be supertrue in all models. So if, in M, B expresses A’s truth-condition,
then B and pA is trueq have the same truth-value at all sharpenings in M because
they have the same content. And if, in M, C expresses A’s falsity-condition, then
C and pA is falseq have the same truth-value at all sharpenings in M because they
have the same content. So if the truth-condition for ‘grass is green’ is that grass is
green, and if the falsity-condition for ‘grass is green’ is that grass is not green, then
any sharpening where either ‘grass is green’ is true or ‘grass is not green’ is true is a
sharpening where ‘ ‘grass is green’ is true ∨ ‘grass is green’ is false’ is true. By LEM:
every sharpening is one where either ‘grass is green’ is true or ‘grass is not green’
is true. So every sharpening is one where ‘ ‘grass is green’ is true ∨ ‘grass is green’
is false’ is true. So ‘grass is green’ is Bivalent. The argument by cases from TC and
FC to Bivalence is therefore |=global-valid. The Supertruth theorist must resist its
premisses.
The only premisses were (i) TC, (ii) FC, (iii) the truth-condition for ‘grass is
green’ is that grass is green, and (iv) the falsity-condition for ‘grass is green’ is that
grass is not green. Since TC and FC are components of the Supertruth theorist’s
account of sentence-content, they must reject (iii) or (iv). Symmetry suggests they
will reject both. But what replacements will they offer? Since the analysis of clarity
in terms of truth is the source of this commitment, that analysis is incomplete (since
it has barely even begun) until replacements are supplied. No Supertruth theorist
has yet done so.
Given the fundamental nature of truth, it is prima façie doubtful whether the
Supertruth theorist can supply alternatives to ‘grass is green’ and ‘grass is not
green’ as truth- and falsity-conditions for ‘grass is green’. Thus it is doubtful
34 The Supertruth theorist denies that mutual entailment is sufficient for sharing of truth-
conditional content. Only a |=global-valid material biconditional suffices for identity of truth-value at
all sharpenings. Since the Supertruth View cashes out content via distribution of truth-value across a
space of sharpenings, only a |=global-valid material biconditional suffices for identity of content. Thus
if A expresses the truth-condition of B, then |=global A↔ (B is true), provided we restrict models to
those that respect either the intended senses of A and B, or the intended relationship between their
senses; including an axiom for ‘is true’ as a base clause in the recursive definition of enforces this
second restriction.
Supervaluations 108
whether the Supertruth theorist can explain their conception of truth. Should we
expect them to be able do so? The demand for an account of the truth-conditions of
‘grass is green’ is a demand for a sentence whose content is exhausted by one of the
two conditions that together comprise the content of ‘grass is green’. How might
such a sentence enter our language? Not via the same route as ‘grass is green’; for
then the two would be vague in just the same ways. The most we have any right to
expect, the Supertruth theorist may claim, is an infinite disjunction, each of whose
disjuncts describes a possible state sufficient for grass to be green (and there seems
little reason to expect even that). Unfortunately, no such sentence will be statable.
In order to make it statable, we need a condition φ common to exactly those states
in which one of the disjuncts is satisfied—that is, a condition common to exactly
those states sufficient for grass to be green—so that we can say that one of them
obtains:
‘Grass is green’ is true iff ∃x(φx).
Yet the problem remains: what right have we to expect an English sentence other
than ‘ ‘grass is green’ is true’ or ‘grass is green’ that will be true in exactly those
possibilities where grass is green? And what right have we to expect an English
predicate true of exactly those states sufficient for grass to be green, other than ‘is
a state of grass’s being green’? Yet without such expressions, any account of φ will
be unilluminating.
An inability to state an adequate truth-condition for ‘grass is green’ other than
‘ ‘grass is green’ is true’ is insufficient to refute the Supertruth theorist because they
are not committed to there being any informative account of such a condition. But
it should make us reluctant to endorse the view. For until an appropriate condition
has been informatively specified, we lack guarantee that the Supertruth theorist’s
conception of truth is both contentful and coherent. Since the Supertruth View
explains clarity in terms of truth, we lack guarantee, or even positive reason to
believe, that the proposed explanation of clarity is both contentful and coherent. If
truth is super-truth, then non-standard accounts of truth and falsity are required.
These have not been provided and it is doubtful that they could be. The Supertruth
View thus incurs a possibly un-meetable explanatory obligation.
Supervaluations 109
2.7.1.3 The Sharpening View
This problem does not afflict the Sharpening View because on that view, truth un-
der an interpretation is truth under a classical interpretation. For any sentence A,
model M and sharpening s ∈ M, exactly one of the following holds:
s, M A.
s, M ¬A.
Since A is false iff ¬A is true, the Sharpening view satisfies Bivalence. The follow-
ing axiom for ‘is true’ ensures that A and pA is trueq receive the same truth-value
at every sharpening and therefore express the same content under every interpre-
tation that respects the intended sense of ‘is true’:
s, M pA is trueq iff s, M A.
Because the Sharpening theorist doesn’t treat clear truth as a semantic classifica-
tion, but as a partly semantic and partly metasemantic classification, they can en-
dorse any account of truth available to the classical semanticist.
2.7.2 Borderline discretion
§2.3.3.2 observed that we must occasionally decide whether to count a borderline
case as a positive case or a negative case. This commonplace feature of linguistic
usage brings none of the discomfort of misuse: it seems compatible with (and per-
haps even partially constitutive of) competence with the expressions in question.
This section argues that the Supertruth View cannot accommodate this.
2.7.2.1 Supertrtuth and borderline discretion
Let a be a borderline F. Suppose you are in a situation where a decision is required
about whether or not to count a as an F. Either choice is open to you. Because such
situations are commonplace and unremarkable, a maximally satisfactory semantics
would allow for either decision without misclassification. The Supertruth View
cannot do so.
Supervaluations 110
On the Supertruth View, neither Fa nor ¬Fa is true when a is a borderline F. So
to count a as an F or to count it as a non-F, is to misclassify it. If truth is supertruth,
then this feature of our use of vague language is, strictly, a misuse of language. But
since meaning is determined by use, an expression’s semantic properties should be
compatible with most aspects of its use, and certainly with all of its most deeply
entrenched ones. Indeed, the legitimacy of those uses should flow naturally from
the correct semantic theory. If truth is super-truth, then this is not so. The truth-
value gaps used to explain borderline ignorance provide too strong an explanation,
one that renders seemingly legitimate uses of language illegitimate.
2.7.2.2 Sharpenings and borderline discretion
The Sharpening View lessens the problem without dissolving it entirely. On that
view, borderline status is not a semantic status incompatible with truth and incom-
patible with falsity. Instead, the sharpenings provide a range of semantic classifi-
cations, only some of which are incompatible with counting a borderline F as, say,
an F. Doing so still involves misclassification, but it also involves correct classifi-
cation. If we can treat these as cancelling each other out, then a perfectly balanced
borderline case can be counted either way without misclassification (though with-
out correct classification also).
The Sharpening View’s core thesis is that the meaning-determining facts de-
termine many intended interpretations. We’ve just seen that this weakens the ob-
jection from borderline discretion. A natural addition to the view eliminates it
entirely. This addition allows decisions about the classification of borderline cases
to count amongst the meaning-determining facts in such a way that deciding to
count a as, say, an F suffices for interpretations that make Fa false to count as un-
intended, provided a is a borderline F: classificatory decisions about borderline Fs
settle their status with regard to F by narrowing the semantic properties of F.
This approach makes it context-sensitive just which interpretations are intended.
The notion of an intended interpretation ought therefore to be relativised to a par-
ticular conversational context or sub-community of a whole linguistic community;
for otherwise my decision to count a borderline F as an F would affect the legit-
Supervaluations 111
imacy of your decision not to, even if you are in a different town from me. This
relativisation is natural if we think of intended interpretations as encoding the in-
formation communicated by uses of language: those privy to my decision to count
a as an F can recover different information from my uses of Fa than those not privy
to that decision. On this view, the intended interpretations of my and my listener’s
shared language vary across contexts, depending on whether I am communicating
with one group or the other because different groups can recover different infor-
mation from my utterances.
These temporary classificatory decisions shouldn’t affect which things count as
borderline cases: deciding to count a terracotta pot as a red pot doesn’t prevent it
from being borderline red/orange. There are two natural and complementary ways
of achieving this. According to the first, the whole community’s language use de-
termines a range of intended interpretations. These settle the borderline cases and
limit the classificatory decisions available to the community’s members. The sec-
ond approach relativises the notion of a borderline case to a linguistic community
or context. A community’s use of language settles a range of intended interpre-
tations that limit the classificatory decisions available to the members of its sub-
communities. The decisions of these sub-communities c don’t affect the intended
interpretations determined by the linguistic behaviour of the wider community
c∗, and hence don’t affect what counts as borderline relative to c∗ despite affect-
ing what counts as borderline relative to c. Whichever approach we prefer, the
borderline cases of English predicates are invariant across the community of En-
glish speakers, despite the decisions of particular speakers affecting the intended
interpretations of their utterances within the contexts in which those classificatory
decisions are made.
Is a similar response available to the Supertruth theorist? Can’t they also treat
the class of sharpenings as context-sensitive and responsive to our classificatory
decisions about borderline cases? Maybe they can. But there is a difficulty to over-
come first. Context-sensitivity of sharpenings can make it correct to count a bor-
derline F as an F following a decision to do so. This does not however, legitimise
making that initial decision; for that decision was made in a context where the bor-
derline F in question occupied a semantic status incompatible with its being an
Supervaluations 112
F. Decisions to count borderline Fs one way or the other are decisions to count
it as something it is not; they are decisions to mis-classify. The problem for the
Supertruth View isn’t whether we can ultimately judge borderline classificatory
decisions correct once they have been made, but whether we can legitimately make
them in the first place. Nothing similar affects the Sharpening View.
2.7.3 Supervaluationist semantics: concluding remarks
We’ve seen two problems for supervaluationist semantics. One concerned the iden-
tification of truth with supertruth. We saw that this incurs a possibly unsatisfiable
explanatory burden. This doesn’t refute the Supertruth View, but it does (i) create
doubt about whether that view is both contentful and coherent, and (ii) under-
mine the Supertruth theorist’s claim to offer an informative analysis of clarity. This
problem does not afflict the Sharpening View.
The second problem was that the Supertruth View seems unable to accommo-
date the legitimacy of temporary decisions about the classification of borderline
cases. Although such decisions are commonplace, unremarkable and practically
indispensable, the Supertruth View regards them as misuses of the expressions
in question, given their semantic properties. The difficulty was shown to be less
pressing for the Sharpening View.
The following two sections present two further problems for the Supertruth
View.
2.8 Field on truth and super-truth
It seems misguided even to try and discover when was the last second of Bertrand
Russell’s youth.35 Hartry Field objects to the Supertruth theorist’s explanation we
presented in §2.5.1.1:
“The supervaluationist says that at certain stages, Russell was neither in
the determinate positive extension nor the determinate negative exten-
sion of ‘old’. But of what possible interest is this, given that (according
35 Suppose for simplicity that time isn’t dense.
Supervaluations 113
to the view) he was at those stages either old or not old?” (Field, 2008,
p.155)
The Supertruth theorist has a reply. If truth is supertruth, then a predicate’s de-
terminate positive and negative extensions are its positive and negative extensions:
the sets of things of which it is true and of which it is false, respectively. So when
Russell was borderline old, neither ‘Russell is old’ nor ‘Russell is not old’ was true.
Since knowledge implies truth, it is neither knowable that Russell was then old,
nor knowable that he was not.
Field responds:
“[C]onsider the question of why a sentence being indeterminate pre-
cludes our knowing it. Calling indeterminateness “lack of truth value”
might appear to provide an answer: you can’t know what isn’t true,
and if indeterminate sentences lack truth value then you obviously can’t
know them! But this is just more verbal hocus pocus: what underlies
the claim that you can’t know what isn’t true is that you can’t know that
p unless p. You can’t know that Russell was old at n nanoseconds un-
less he was old at n nanoseconds, and you can’t know that he wasn’t
old at n nanoseconds unless he wasn’t old at n nanoseconds. But on the
supervaluationist view he either was or wasn’t, and if you can’t know
which, that needs an explanation. The use of ‘true’ to mean super-true
just serves to disguise this.” (Field, 2008, p.154)
This shouldn’t immediately convince the Supertruth theorist. For she claims that
when Russell was borderline old, the disjunction ‘Russell is old or Russell isn’t old’
was true, but neither disjunct was: it wasn’t the case that Russell was old, and it
wasn’t the case that Russell wasn’t old. This is contradictory if ‘it’s not the case that
Russell is not old’ involves two occurrences of the same kind of negation: ¬¬A.
The Supertruth theorist therefore needs the outer negation to form a truth from
any untruth and the inner one to form a truth only from falsehoods (and form val-
ueless sentences from other valueless ones). This casts doubt on the inner (strong)
negation sign’s claim to express genuine negation. The outer (weak) negation is
what’s used to explain why we can’t know whether Russell was old when he was
Supervaluations 114
borderline old; only in that sense is it not the case that A and not the case that ¬A
when A is valueless. But only the inner (strong) negation is governed by superval-
uationist semantics; for on that semantics ¬A and ¬¬A are both valueless when A
is borderline. The Supertruth View either gives the wrong account of negation or
cannot explain borderline ignorance.
This doesn’t touch the Sharpening View. §2.5.1.1 offered two candidate neces-
sary conditions on the truth of pS knows that Aq under an intended interpretation
s:
The proposition s assigns to A is true.
Each proposition assigned to A by any intended interpretation is true.
Both ensure that pS knows that Aq is no better than borderline when A is border-
line. Given the following rule, it follows that we ought not claim to know that A
when A is borderline:
Assert only the truth.
Borderline status thus makes investigation into known borderline claims misguided
by making it in-principle illegitimate to claim to know the result of the investiga-
tion. Since this doesn’t appeal to the untruth of borderline claims, the Sharpening
View is immune to Field’s objection.
2.9 Higher-order vagueness
We introduced vagueness as the fuzziness characteristic of the red/orange, tall/not
tall and intelligent/unintelligent distinctions. The extent of this fuzziness is itself
fuzzy. This gives rise to the phenomenon of higher-order vagueness. This section
examines some objections to supervaluationist accounts of it.
2.9.1 Terminology
We begin with some terminology. Williamson (1999) develops these ideas more
carefully.
Supervaluations 115
Consider the classification of objects into the Fs and the non-Fs. Call this
the zero-order F-classification. Objects may clearly belong to one of its sub-
classifications. F is first-order vague iff there could be objects that don’t clearly
belong to any sub-classification of the zero-order F-classification; iff there could be
borderline cases to the zero-order F-classification. Such first-order borderline cases
of F are neither clearly F nor clearly not F.
Consider this last classification into the clear Fs, first-order borderline Fs and
clear non-Fs. Call this the first-order F-classification. Objects may clearly belong
to one of its sub-classifications. F is second-order vague iff there could be objects
that don’t clearly belong to any sub-classification of the first-order F-classification;
iff there could be borderline cases to the first-order F-classification. Such second-
order borderline cases of F are either:
(i) neither clearly clearly F nor clearly not clearly F; or
(ii) neither clearly first-order borderline F nor clearly not first-order borderline
F; or
(iii) neither clearly clearly not F nor clearly not clearly not F.
Consider this last classification into the (i) clearly clear Fs, (ii) borderline cases
of clear Fs, (iii) clearly first-order borderline Fs, (iv) borderline cases of first-order
borderline Fs, (v) clearly clearly not Fs, and (vi) borderline cases of clearly not Fs.
Call this the second-order F-classification. Objects may clearly belong to one of
its sub-classifications (i)–(vi). F is third-order vague iff there could be objects that
don’t clearly belong to any sub-classification of the second-order F-classification;
iff there could be borderline cases to the second-order F-classification. We won’t
list the possibilities for these third-order borderline Fs.
Iterating this construction allows us to define arbitrarily high orders of border-
line case and vagueness. We can extend it to borderline sentences via the stipulation
that an ith-order borderline sentence is an ith-order borderline case of a truth. F
is precise iff F is not ith-order vague, for any i > 0. F is higher-order vague iff F is
ith-order vague for some i > 1.
Supervaluations 116
2.9.2 Varieties of higher-order vagueness
This section examines the relationship between the technical notions defined in the
previous section and the phenomenon they are intended to capture.
The arguments for first-order vagueness extend naturally to higher-order vague-
ness. There seems no non-arbitrary stopping point. It might be objected that the
world itself may not be fine-grained enough to allow distinctions between every
definable order of borderline case, and hence that, above some level, the orders
collapse into one. This may be right. But this collapse is imposed by the world our
vague concepts describe, rather than by those concepts themselves. An adequate
analysis of vagueness ought not to presuppose it.
Another objection to arbitrarily high orders of vagueness is that it rapidly out-
strips our capacity to comprehend; understanding attributions of third-order vague-
ness is, for most, a very difficult task. But difficulties with understanding don’t
imply non-existence. The case is similar to arbitrarily long and complex sentences
of English. They may not be comprehensible, but they are still meaningful. Roy
Sorensen’s (2010) contains related discussion and arguments for arbitrarily high
orders of vagueness.
These considerations motivate the thesis of:
Unrestricted Borderline Cases (UBC) If F is vague, then there could be border-
line cases to any sub-classification of the ith-order F-classification, for all i.
There are no limits on higher-order vagueness.36
Compatibility with UBC is an attractive feature of theories of vagueness. Note
however that this technical notion is intended to capture an intuitive idea of in-
eradicable fuzziness:
No Sharp Boundaries (NSB) Vague predicates impose only fuzzy classifications,
they mark no sharp boundaries whatsoever.
What is the relationship between UBC and NSB? There are two attitudes one might
take.36 UBC is stronger than the following: vague predicates are ith-order vague, for all i. UBC re-
quires borderline cases to every sub-classification of the ith-order classification, while this weaker
alternative only requires borderline cases to some sub-classification of the ith-order classification.
Supervaluations 117
The first approach takes UBC as an analysis of NSB. This view identifies fuzzi-
ness with the (possible) presence of borderline cases. The ineradicable fuzziness of
NSB is then identified with there being no restrictions on higher-order vagueness.
The alternative view takes UBC as a consequence of NSB, but not as an anal-
ysis. On this view, ineradicable fuzziness is responsible for the existence of unre-
strictedly high orders of borderline case, though the content of the former notion
may outstrip that of the latter. We might even add that the fuzziness of the initial
F/non-F classification is what’s responsible for NSB: to be fuzzy is to be inerad-
icably fuzzy, and hence to entirely lack sharp classificatory boundaries. On this
view, the orders of vagueness are manifestations of the underlying phenomenon of
fuzziness.
It shouldn’t matter to our discussion which approach is correct. We’ll focus on
UBC. Since it’s very hard to see how a classification could be fuzzy and yet not allow
for borderline cases, UBC seems necessary for NSB. Subsequent sections examine
two arguments against UBC.
2.9.3 Higher-order vagueness and the Supertruth View
On the Supertruth View, if A is first-order borderline, then it lacks truth-value.
What about if A is second-order borderline? We have three options: true, val-
ueless, and false. The second-order borderline cases therefore collapse into the
same semantic status as either the clear cases, the first-order borderline cases or
the clear non-cases. Higher-order vagueness is not distinctive at the level of truth-
evaluation.
The Supertruth theorist might respond by introducing more truth-values. This
brings three problems: (i) it complicates the theory; (ii) the complication is severe
because UBC implies that there are infinitely many orders of borderline case, and
hence also truth-values; (iii) each extra truth-value requires philosophical explana-
tion.37 An alternative would be preferable.
This suggests that if truth is super-truth, then higher-order vagueness in F is
not a feature of the original F/non-F classification:
37 This explanatory burden is new. Our Supertruth View employs truth-value gaps in place of a
third value.
Supervaluations 118
“It may be misleading to think of higher-order vagueness in α as a
species of vagueness in α. Higher-order vagueness in α is first-order
vagueness in certain sentences containing α.” (Williamson, 1999, p.140)
Focus on the clear end of a Sorites series. The second-order borderline sentences
have the same truth-status as either the clear or the first-order borderline sentences.
The difference emerges in the truth-status of sentences containing them: ∆A is true
when A is clearly true, false when A is first-order borderline, and valueless when
A is second-order borderline. Let’s accommodate this formally.
2.9.4 Semantics for higher-order vagueness
We want to complicate the supervaluationist formalism to allow distinctions amongst
varieties of borderline case. We begin with second-order borderline cases, then
third-order borderline cases, and then arbitrarily high-ordered borderline cases.
Our strategy iterates the supervaluationist construction to allow distinctions within
a model-structure amongst the sentences that fall down a truth-value gap.
Remove ∆ from the object-language. We’ll replace it with something more ad-
equate shortly. Supervaluationist models are re-named 1-models. A 2-model M2
is a class of 1-models. Our original base clauses are amended with an additional
relativisation of to 2-models:
v, s, M1, M2 Φnα1, . . . , αn iff 〈Jα1Ks,v, . . . , JαnKs,v〉 ∈ JΦnKs.
v, s, M1, M2 ¬A iff v, s, M1, M2 1 A.
v, s, M1, M2 A ∧ B iff v, s, M1, M2 A and v, s, M1, M2 B.
v, s, M1, M2 ∀xA iff v′, s, M1, M2 A, for every assignment v′ that differs
from v at most over ‘x’.
s, M1, M2 A iff v, s, M1, M2 A, for all assignments v.
Define supertruth and superfalsity in a 2-model:
A is supertrue in M2 iff s, M1, M2 A, for all 1-models M1 ∈ M2 and sharp-
enings s ∈ M1.
Supervaluations 119
A is superfalse in M2 iff s, M1, M2 1 A, for all 1-models M1 ∈ M2 and sharp-
enings s ∈ M1.
Supertruth (superfalsity) in a 2-model is supertruth (superfalsity) in all of its 1-
models. The Supertruth View now identifies truth under an interpretation with
supertruth in a 2-model, and falsity under an interpretation with superfalsity in
a 2-model. Plugging this into the Tarskian analysis of consequence gives a new
account of global consequence:
Γ |=global C iff, for every 2-model M2, if every member of Γ is supertrue in
M2, then C is supertrue in M2.
To express claims about first-order vagueness, we add a sentential operator ∆1:
v, s, M1, M2 ∆1A iff v, t, M1, M2 A, for all sharpenings t ∈ M1.
Then we add another operator ∆2 for expressing claims about second-order vague-
ness:
v, s, M1, M2 ∆2A iff v, s, N1, M2 A, for all 1-models N1 ∈ M2.
With ∆1 and ∆2 in place, first-order borderline cases are distinguishable from second-
order borderline cases.
Third-order borderline cases are accommodated by a further iteration. A 3-
model is a class of 2-models. Further relativise the base clauses to 3-models. Define
supertruth and superfalsity in a 3-model:
A is supertrue in M3 iff s, M1, M2, M3 A, for all 2-models M2 ∈ M3, 1-
models M1 ∈ M2 and sharpenings s ∈ M1.
A is superfalse in M3 iff s, M1, M2, M3 1 A, for all 2-models M2 ∈ M3, 1-
models M1 ∈ M2 and sharpenings s ∈ M1.
Truth and falsity under an interpretation are then identified with supertruth and
superfalsity in a 3-model. Consequence becomes supertruth-preservation in every
3-model. Finally, a ∆3 operator is introduced for expressing claims about third-
order borderline cases:
v, s, M1, M2, M3 ∆3A iff v, s, M1, N2, M3 A, for all 2-models N2 ∈ M3.
Supervaluations 120
And so on upwards. The construction can be iterated indefinitely to capture indef-
initely high orders of vagueness. The general forms of the rules for supertruth and
superfalsity in an i-model, and for each ∆i operator are:
A is supertrue in Mi iff s, M1, . . . , Mi A, for all (i− 1)-models Mi−1 ∈ Mi,
(i− 2)-models Mi−2 ∈ Mi−1,. . . , and sharpenings s ∈ M1.
A is superfalse in Mi iff s, M1, . . . , Mi 1 A, for all (i− 1)-models Mi−1 ∈ Mi,
(i− 2)-models Mi−2 ∈ Mi−1, . . . , and sharpenings s ∈ M1.
v, s, M1, . . . , Mi ∆i A iff v, s, M1, . . . , Mi−1, Mi A, for all (i − 1)-models
Mi−1 ∈ Mi.
Two closing comments. Firstly, if ∆i A is supertrue in an i-model, then so is
∆i−1A, as it should be. Secondly, falsity at any sharpening suffices for untruth at
any 1-model containing it; which suffices for untruth at any 2-model containing
that 1-model; which suffices for untruth at any 3-model containing that 2-model;
which suffices. . . . The merest hint of unclarity suffices for untruth. Different orders
of borderline case are distinguished not by their semantic relationship to F, but to
open sentences containing F, e.g.: ∆1Fx.
2.9.5 Hidden sharp boundaries?
The following are supertrue in any 1-model:
∆1A→ ∆1∆1A(S4)
¬∆1A→ ∆1¬∆1A(S5)
It might seem that this rules out higher-order vagueness: neither the clear cases
nor the less-than-clear cases can have borderline cases. But ∆1 is only intended to
express claims about first-order borderline cases, not higher-order ones. Iteration
of ∆1 is an artefact of the formation rules without representational import.
It would be bad news if the following were supertrue in any i-model:
∆i−1A→ ∆i∆i−1A(S4i)
¬∆i−1A→ ∆i¬∆i−1A(S5i)
Supervaluations 121
(S4i) rules out borderline cases to the ∆i−1 cases. (S5i) rules out borderline cases
to the less than ∆i−1 cases. Were either supertrue on all i-models, such kinds of
borderline case would be logically impossible and UBC would be false. Fortunately,
both fail. Let M2 be a 2-model containing only the 1-models M1, N1 such that:
(2) s, M1, M2 A, for all s ∈ M1
and:
(3) s, N1, M2 1 A, for all s ∈ N1
We now show that neither (S4i) nor (S5i) is supertrue in M2 (for i = 2). Variables
ranging over sharpenings are treated as implicitly universally quantified to help
with presentation.
From (3) and the rule for ∆1: s, N1, M2 1 ∆1A. So by the rule for ∆2: s, M1, M2 1
∆2∆1A. But from (2) and the rule for ∆1: s, M1, M2 ∆1A. Instantiating (S4i) for
i = 2 therefore gives a conditional ∆1A → ∆2∆1A whose antecedent is true at
s, M1, M2 and whose consequent is not. Hence: s, M1, M2 1 ∆1A → ∆2∆1A. So
(S4i) is not supertrue in M2 (for i = 2).38 The argument generalises to show that
for no i is (S4i) supertrue on all i-models.
From (2) and the rules for ∆1 and ¬: s, M1, M2 1 ¬∆1A. So by the rule for ∆2:
s, N1, M2 1 ∆2¬∆1A. But from (3) and the rules for ∆1 and ¬: s, N1, M2 ¬∆1A.
Instantiating (S5i) for i = 2 therefore gives a conditional ¬∆1A → ∆2¬∆1A whose
antecedent is true at s, N1, M2 and whose consequent is not. Hence: s, N1, M2 1
¬∆1A → ∆2¬∆1A. So (S5i) is not supertrue in M2 (for i = 2). The argument
generalises to show that for no i is (S5i) supertrue on all i-models.
Williamson (1994, §5.6) defines an operator ∆∗ as equivalent to an infinite con-
junction:
∆∗A is true iff ∆1A is true, and ∆2∆1A is true, and ∆3∆2∆1A is true, and. . .
38 Note that (S4i) is not superfalse in M2. From (3): s, N1, M2 1 ∆1 A. So: s, N1, M2 ∆1 A →
∆2∆1 A. So (S4i) is not superfalse in M2 (for i = 2). Since the argument in the text shows that it
isn’t supertrue either: (S4i) falls down a supertruth-value gap in M2 (for i = 2). In fact, (S4i) is not
superfalse in any i-model. Similar remarks apply to (S5i).
Supervaluations 122
The ∆∗ cases represent the maximally clear cases: those without a hint of vague-
ness. Williamson claims that the following is valid:
∆∗A→ ∆∗∆∗A(S4*)
He concludes that the maximally clear cases are sharply bounded. If so, then the
Supertruth View places a logical limit on the extent of higher-order vagueness:
borderline ∆∗ cases are logically impossible. So UBC, and hence NSB, are false. But
once one sharp boundary is accepted, what’s wrong with more? Why is a sharp
distinction between the ∆∗ cases and the rest better than one between the cases
and the non-cases? In other words: why not adopt an epistemic account of all
vagueness, given that we have to do so for vagueness in ∆∗? Furthermore, since
even a hint of unclarity suffices for untruth, the positive cases will be the ∆∗ cases,
and hence sharply distinguished from the rest. So there cannot really even be first-
order borderline cases. The Supertruth View looks highly unstable.
Williamson’s argument is not irresistible. Note first that the clause for ∆∗, un-
like those for our ∆i, employs a notion of truth without relativisation to any kind
of model-structure. Thus ∆∗ isn’t well-defined in our framework. How can this be
rectified? The most promising strategy combines 1-models, 2-models, 3-models,
and i-models for every natural i into a single structure of the kind defined in the
previous section. Truth in that kind of structure can then be used to supply truth-
conditions for ∆∗:
v, s, M1, M2, M3 . . . ∆∗A iff:
(i) v, t, M1, M2, M3 . . . A, for all t ∈ M1, and
(ii) v, s, N1, M2, M3 . . . A, for all 1-models N1 ∈ M1, and
(iii) v, s, M1, N2, M3 . . . A, for all 2-models N2 ∈ M3, and...
This validates (S4*). Does it show that ∆∗ is precise, or that supervaluationist se-
mantics imposes hidden sharp boundaries? A positive answer requires (a) that
vagueness does not extend into transfinite orders, and (b) that this kind of model-
structure can capture all the vagueness of a natural language with only finite orders
of vagueness.
Supervaluations 123
Set aside objections to (a): if each finite order of vagueness is captured by some
iteration of our supervaluationist construction, then transfinite orders of vagueness
should be captured by transfinite iterations. (Shapiro, 2006, ch.5.1 contains related
discussion.) And even if not, the Supertruth theorist surely shouldn’t have to appeal
to something so recherché as transfinite orders of vagueness.
Assumption (b) is more dubious, and certainly not mandatory. We presented
supervaluationism as a reasonable mathematical approximation to vague classifi-
cation. The assumption that all the vagueness of a natural language can be cap-
tured without artefacts by a single mathematical structure is non-trivial. In fact,
there are reasons independent of supervaluationism to doubt that it can be, and to
which we now turn.
2.9.6 Sainsbury on vagueness and set-theoretic semantics
Mark Sainsbury argues thus:
“Sets have sharp boundaries, or, if you prefer, are sharp objects: for any
set, and any object, either the object quite definitely belongs to the set
or else it quite definitely does not. Suppose there were a set of things of
which “red” is true: it would be the set of red things. However, “red”
is vague: there are objects of which it is neither the case that “red”
is (definitely) true nor the case that “red” is definitely not true. Such
an object would neither definitely belong to the set of red things nor
definitely fail to belong to this set. But this is impossible, by the very
nature of sets. Hence there is no set of red things.” (Sainsbury, 1990,
p.252)
For similar reasons, there can be no set of clearly red things, or clearly clearly red
things, and so on. Granting Sainsbury’s assumption about the sharpness of sets, it
follows that no set-theoretic semantics can capture the vagueness of natural lan-
guage. And if all mathematics can be captured within set-theory, then no mathe-
matised semantics can capture vague classification without inaccuracy. Vagueness
proper will be just what is missing from any such semantics; only a formal surro-
gate can remain.
Supervaluations 124
This line of thought is attractive. But what exactly does it show? It does not
show that mathematised semantics cannot provide insight into vague classifica-
tion, only that it cannot exhaust vagueness. It remains open whether a given math-
ematical structure might closely resemble that of an appropriately circumscribed
segment of vague classification. And that is all the Supertruth theorist need claim.
So long as our theoretical interest lies only in the structure of the F/non-F clas-
sification, we can rest content with considering individual sharpenings. Each will
classify—misclassify, the Supertruth theorist will claim—some borderline cases one
way or the other. But if we aren’t interested in such close approximation—in distin-
guishing the clear from the borderline, and in assigning truth-values only to those
(contentful) sentences that possess them—this need not undermine our employing
a semantic theory based around individual sharpenings. Unless we are interested
in vagueness, we need not require our semantic theory to respect, or even be capa-
ble of representing, vagueness-related truth-value gaps.
An interest in first-order borderline cases requires a different approach, based
around 1-models and ∆1. This improves on the classical semantic theory based
around individual sharpenings by allowing expression of the claim that x is a
borderline F, and thereby distinguishing the first-order borderline from the clear
cases.
An interest in second-order vagueness requires a different approach again. The
distinctions afforded by individual 1-models are too coarse for this. We need to
consider 2-models and introduce ∆2. This allows us to distinguish between cases,
first-order borderline cases and second-order borderline cases. Think of this as an
open-ended process. Moving up through new kinds of model provides better and
better approximations, each capable of representing more aspects of vague classifi-
cation than its predecessor. But we should not assume without argument that even
the limit of this process will be perfectly accurate. And Sainsbury’s argument pro-
vides positive reason to doubt that it will be, though without undermining the use-
fulness of supervaluationist semantics, provided its limitations are kept in mind.
On this approach, higher-order vagueness is analogous to indefinite extensibil-
ity:
Supervaluations 125
F is indefinitely extensible iff there is a function δ such that, for any collection
x of Fs, δ(x) is an F that does not belong to x.
Although this lacks definite content without explication of the relevant (and some-
what murky) notions of function and collection, it will suffice for our purposes.
Indefinitely extensible concepts are supposed to resist the formation of a collection
that exhausts their instances: for any collection x, purported to be the collection of
all Fs, δ(x) is an F that’s not amongst x; hence x is not the collection of all Fs.
Likewise, a higher-order vague concept resists complete characterisation of its
applicability and vagueness. Suppose we attempt to describe the applicability of
vague F. A simple demarcation into the Fs and non-Fs is, at best, only borderline
correct because it classifies some borderline cases one way or the other. So we intro-
duce the notion of clarity. The sole purpose of this notion is to delimit the source
of the original description’s borderline status. But if F is second-order vague, the
resulting description will also be only borderline correct because it counts some
borderline clear Fs as clear Fs (for example). Unrestricted higher-order vague-
ness in F prevents complete description of the applicability and vagueness of F.
Each attempted description can be no better than borderline correct. Exhausting
the vagueness of F requires a clearly correct description of the ways in which all
descriptions are only borderline correct. But since higher-order vagueness pre-
vents any description from capturing the ways in which it is itself only borderline
correct, it’s impossible to exhaustively describe the vagueness of F. (This meshes
nicely with the view on which the ineradicable fuzziness of NSB implies, but is not
analysed by, the higher-order borderline cases of UBC.)
If this line of thought is correct, then we may have an explanation of why no
set-theoretic structure is vague: the nature of a set is exhausted by a list of its mem-
bers, but no list of its instances can exhaust the applicability of a vague predicate,
regardless of how fine-grained the distinctions we make amongst items on the list.
The present approach differs from a prominent alternative: insist that an ade-
quate specification of the sharpenings requires a vague metalanguage (Keefe, 2000,
ch.8 §1). Vagueness in the specification of sharpenings can induce vagueness in
whether a sentence is true at them all, and hence vagueness in the truth-status of
Supervaluations 126
claims about clarity. Our approach, by contrast, makes no appeal to a vague met-
alanguage. Instead, we accept an in-principle limit on how closely our model can
approximate vague classification. Our Supertruth theorist’s metalanguage is the
standard (and precise) language of classical mathematics. This affords a significant
advantage over the alternative: we do not need to know what forms of reasoning
are valid in a vague language before our investigation begins. Were our metalan-
guage vague, we would need to know the effect of vagueness on validity before we
could derive any results about the model. But studying validity is just what we
want the model for. So we use a standard mathematical metalanguage to approxi-
mate vagueness as best we can.
2.9.7 Objection: the fragmentation of vagueness
The present treatment distinguishes ∆1 and ∆2 by their semantic axioms and the
kinds of structure those axioms presuppose. This section considers the objection
that this misrepresents the unitary nature of clarity: our proposal on behalf of
the Supertruth theorist breaks clarity into a cluster of formally similar distinct
concepts, thereby misrepresenting vagueness as a non-uniform phenomenon.
This is not compelling. We distinguished different orders of borderline case
and the kinds of structure needed to represent those orders. But within any model
of any orders of vagueness, vagueness is represented by the structure as a whole,
not any particular component of it. We could even, if we wished, define a single ∆
operator capable of capturing all the orders of vagueness represented by a single
model-structure.39
Furthermore, our approach is formally equivalent to a more common one, against
which the objection is without force. This alternative introduces an accessibility
relation on a single space of sharpenings and characterises a single clarity oper-
ator via truth at all accessible sharpenings, instead of at all sharpenings. Both
approaches impose a hierarchical structure on a space of sharpenings and treat
higher-orders of vagueness in terms of higher levels in the hierarchy. On the alter-
39 Whether truth-value across sharpenings, 1-models, 2-models and so on was relevant to the truth
of a sentence featuring this operator would depend on how many other occurrences of it occurred
within its scope.
Supervaluations 127
native however, there is no temptation to treat clarity as non-uniform. Given this
formal equivalence, that temptation should not arise on our approach either.
Given this equivalence and the alternative’s greater elegance, why bother with
our approach at all? The answer is that it has a philosophical benefit that the al-
ternative lacks. On our approach, consideration of higher and higher orders of
vagueness requires that we consider different kinds of structure and define our se-
mantic axioms anew for each one. Although inconvenient, this reminder of our
theory’s representational limit serves as a warning against assuming the meaning-
fulness of operators like ∆∗, which assume a complete hierarchy of orders within
a single model. It also warns against assuming the possibility of capturing all the
vagueness of natural language within a single mathematical structure; it warns
against assuming that clarity in our representation is always indicative of clarity in
the system it represents.
2.9.8 More hidden sharp boundaries?
Shapiro (2006, p.128) presents an argument similar to Williamson’s ∆∗ argument,
but that makes no explicit assumptions about the semantics of vagueness and is
therefore immune to our response to Williamson’s argument.
Consider the absolutely clear F’s: the F’s about whose F-ness there isn’t even
the slightest hint of unclarity. Absolute clarity is the informal analogue of ∆∗.
Suppose that a is borderline absolutely clearly F. Then there is a hint of unclarity
about a’s F-ness. So a is not absolutely F. But if a is not absolutely F, then it’s not
borderline whether a is absolutely F. Since a was arbitrary and this rests on no
assumptions: borderline absolutely clear F’s are impossible; the absolute F’s must
be sharply bounded.
This is not uncontroversially valid. Let’s use abs for an absolute clarity operator.
The argument began by supposing that a is borderline absolutely clearly F:
¬∆ abs Fa ∧ ¬∆¬ abs Fa(4)
The task is to show that borderline absolutely clear Fs are impossible, and hence
that (4) is false. Shapiro’s argument begins by inferring from (4) that a is not abso-
Supervaluations 128
lutely clearly F on the grounds that there is a hint of unclarity about its F-ness:
¬ abs Fa(5)
The conjunction of (4) and (5) is not a contradiction. So we can’t yet conclude that
(4) is false, or that a is not a borderline absolutely clear F:
¬(¬∆ abs Fa ∧ ¬∆¬ abs Fa)
How might we get a contradiction? Consider a version of the S5 axiom for abs:
¬ abs A→ abs¬ abs A(S5abs)
From (5), this yields:
abs¬ abs Fa
Then because abs A implies ∆A:
∆¬ abs Fa
This contradicts the second conjunct of (4). But we’ve already seen that principles
like (S5abs) shouldn’t be unrestrictedly valid, if we’re going to allow for higher-
order vagueness.
An alternative strategy appeals to: A |=global ∆A. From (5), this yields
(6) ∆¬ abs Fa
Which again contradicts the second conjunct of (4). But A |=global ∆A requires
only that ∆A is (super)true in any model where A is (super)true. This is silent
about the (super)truth-status of ∆A when A is borderline and hence untrue. It’s
therefore silent about the (super)truth-status of (6) under the supposition that (5)
is borderline. Yet that’s just what (4) implies (since ¬A is borderline whenever A
is borderline). So on the Supertruth View, it follows only that if it’s (super)true,
and hence clearly true, that a is not absolutely clearly F, then its’s not borderline
whether a is absolutely clearly F. This is obviously unhelpful when trying to reduce
the supposition that a is borderline absolutely clearly F to absurdity.
This response is problematic. abs Fa is false if (5) is true. But (4) says that abs Fa
is borderline, and hence valueless. Since nothing can be both false and valueless,
Supervaluations 129
(4) and (5) cannot both be true. So (4) cannot be true, if it implies (5). So borderline
absolutely clear cases are impossible after all.
The Supertruth theorist must therefore reject the initial step from (4) to (5).
Although (4) implies that abs Fa isn’t true, it’s negation need not be true: abs Fa
can fail to be true by being borderline, just as (4) says, without thereby being false,
as (5) says.
Given the Supertruth View’s connection between borderline status and truth-
value gaps, the argument from pBorderline-Aq to pNot-Aq shouldn’t be valid.
Shapiro’s argument therefore does not show that borderline absolutely clear cases
are impossible. It does show that all borderline absolutely clear cases fail to be
absolutely clear cases. But that’s compatible with their failing to be non-cases of
absolutely clear cases too, given the identification of borderline status with truth-
value gaps. If an object can satisfy neither A nor ¬A, as the Supertruth theorist
claims, there seems no reason why it couldn’t satisfy neither abs A nor ¬ abs A.
Yet the Supertruth theorist must provide an account of clarity that explains the
compatibility of failure to be a case with being a borderline case, as opposed to
a non-case. Since the borderline cases fall down a truth-value gap, this requires
an account of truth that distinguishes untruth from falsity. Without that account,
the present line of resistance to Shapiro’s argument looks more like wishful think-
ing than a principled response. Since that explanation is just what the Supertruth
theorist has yet to provide—recall the discussion in §2.7.1—their position is tenu-
ous. The difficulty explicating their non-Bivalent conception of truth undermines
the Supertruth theorist’s ability to respond to Shapiro’s argument for hidden sharp
boundaries.
A better supervaluationist strategy would be to find an alternative response.
There seem to be three options. (i) Deny that the concept of absolute clarity is co-
herent. I can see no good reason to grant this. (ii) Accept this limit on higher-order
vagueness, but resist positing sharp boundaries elsewhere. I don’t know how to do
so in a principled manner.40 (iii) Deny that there are any absolutely clear cases. The
40 One strategy might claim that absolute clarity is a theoretical concept and so deny that the
intuitive reasons to acknowledge borderline cases apply to it: the absolutely clear F’s are not sorites-
susceptible because there’s no extra-theoretical motivation to grant a sorites premiss for ‘absolutely
Supervaluations 130
next section argues that the Sharpening View can motivate this denial. I know of
no other way to do so. But without such motivation, response (iii) looks objection-
ably ad-hoc. Hence it is doubtful whether the Supertruth View can accommodate
unrestricted higher-order vagueness.
2.9.9 Higher-order vagueness and the Sharpening View
The Sharpening View fares better with higher-order vagueness. The view is moti-
vated by a picture of Reality as a gradual place; vagueness arises when we impose
non-gradual classifications upon it. This gradualness and the relative coarseness of
the meaning-determining facts combine to ensure that no one classificatory bound-
ary is privileged over all others, despite many being ruled out. Each remaining
boundary provides an intended interpretation of the expression in question.
We can apply this to metasemantic vocabulary. The result is a well-motivated
denial that there are any (typical) absolutely clear or ∆∗ cases. Hence neither
Williamson nor Shapiro’s argument to show that such cases are sharply bounded
shows that actual vague classification is sharply bounded.
Before we begin, it’s worth responding to the following objection: surely there
are absolutely clear cases; isn’t scarlet as clearly a shade of red as anything could
be? The approach below attempts to offset the strangeness of the claim that scarlet
isn’t absolutely clearly a shade of red by allowing that it’s as clearly a shade of red as
anything could be, given our limitations and the way we use language. Extremely
clear cases are commonplace and susceptible to borderline cases, but absolutely
clear cases are not. The Sharpening theorist’s response is to accuse the objector of
confusing absolute clarity with very high levels of clarity.
2.9.9.1 Metasemantic gradualness
We want metasemantic vocabulary to be vague. So we need to show how the facts
we describe using that vocabulary can be gradual, just like the facts described by
typical vague vocabulary.
clear F’ or ‘not absolutely clear F’. But although absolute clarity is a theoretical concept, it doesn’t
seem so far divorced from ordinary clarity that we shouldn’t find attractive either Sorites premisses
for it, or the claim that it permits borderline cases. Thanks to Will Bynoe for suggesting this strategy.
Supervaluations 131
The metasemantic facts impose an ordering on interpretations according to how
well they fit a community’s linguistic behaviour (§2.4.2). For example, an interpre-
tation that places the tall/non-tall distinction at 5’11” fits our use of ‘tall’ better
than one that places it at 5’10”, but (perhaps) less well than one that places it at
6’. Small differences between heights bring small differences in how well interpre-
tations that locate the tall/non-tall boundary at those heights fit our use of ‘tall’.
Gradualness in the height-facts underlying our use of ‘tall’ thus induces gradual-
ness in the metasemantic facts underlying our use of ‘intended interpretation’ (and
vagueness in object-language reflections thereof, like ‘said that’).
This gives an ordering on interpretations according to how well they fit a com-
munity’s use of language. Intended interpretations are greatest elements in this
ordering. But does the class of such greatest elements provide a significantly better
interpretation of ‘intended interpretation’ than any other? Or, like the distinc-
tion between interpretations of ‘tall’ that place the tall/non-tall boundary at 6’
and those that place it at 6’0.00001”, is this a distinction without a difference? In
the former case, metasemantic vocabulary will be non-vague. In the latter case,
metasemantic vagueness seems likely: no one point in the gradual fit-transition
described by non-gradual metasemantic vocabulary is significantly better than all
others. Answering this question requires an account of the nature of the facts un-
derlying discourse about intended interpretations.
Distinguish two broad approaches to metasemantics. One sees metasemantic
facts as a sui generis kind of fact, though systematically connected to other kinds of
fact, such as those about language-use. On this view, the intended/unintended dis-
tinction is naturally taken to mark a significant difference.41 This isn’t threatened
by the fact that the intended/unintended distinction isn’t revealed in a descrip-
tion of the fit-ordering alone; for that distinction isn’t supposed to be reducible to
facts about fit, despite coinciding with the greatest/not-greatest distinction in the
fit-ordering.
The alternative approach sees metasemantic discourse as codifying other facts
41 A conception of semantic facts as sui generis only makes it natural, not mandatory, to regard the
intended/unintended distinction as significant. For that distinction could be a sui generis gradual
distinction. It’s hard to find a motivation for this view.
Supervaluations 132
about, say, the relevance of various propositional contents to linguistic communi-
cation within a community. This kind of approach undermines the significance of
the intended/unintended distinction by denigrating the greatest/not-greatest dis-
tinction in the fit-ordering; although that distinction exists, it marks no significant
difference. What really matters is not which interpretations are greatest in the fit-
ordering, but the ordering itself.
On this second approach, the class of greatest interpretations in the fit-ordering
need not provide a significantly better extension for ‘intended interpretation’ than
a more inclusive class containing some marginally less well fitting interpretations.
Metasemantic theorising mandates drawing a distinction somewhere in the fit-
ordering, though no one candidate is significantly better than all others. Vagueness
can then infect metasemantic concepts just as it infects any others. In this case, it
can be vague which sentences are clearly true.
This suggests a philosophical interpretation of the modified supervaluationist
formalism described in §2.9.4. Sharpenings represent interpretations of the non-
metasemantic vocabulary. 1-models represent classes of such interpretations: in-
terpretations of ‘intended interpretation’. 2-models thus represent classes of inter-
pretations of ‘intended interpretation’. By parity of reasoning, 2-models represent
interpretations of ‘intended interpretation of ‘intended interpretation’ ’; they rep-
resent states of the metasemantic facts on which there are many intended interpre-
tations of ‘intended interpretation of the non-metasemantic vocabulary’. Likewise
mutatis mutandis for 3-models and above.
Think of ‘intended interpretation’ as marking a threshold in the fit-ordering:
interpretations that fit better than the threshold count as intended. Objects that
satisfy F under each interpretation that exceeds this threshold are clearly F (un-
der that interpretation of ‘intended interpretation’). Thresholds in the fit-ordering
correspond to 1-models.
Now, if the metasemantic facts are gradual—i.e. if nearby interpretations in
the fit-ordering fit the meaning-determining facts almost as well as each other—
then many nearby thresholds will have equal claim to be the threshold marked by
‘intended interpretation’: none marks any significant difference in how well inter-
pretations that exceed that threshold fit the meaning-determining facts, or their
Supervaluations 133
relevance to linguistic communication within the community in question. Objects
that satisfy F under each interpretation that meets each of these thresholds are
clearly clearly F. Classes of thresholds in the fit-ordering correspond to 2-models.
The class of interpretations that meet any of these thresholds corresponds to the
class of sharpenings that belong to any of the 1-models within a 2-model.
Provided the metasemantic facts are sufficiently gradual, this process should it-
erate: many classes of classes thresholds in the fit-ordering will have equally good
claim to be the intended interpretation of ‘intended interpretation of ‘intended
interpretation’ ’, and so on. The key point is that iterating ‘clearly’ to force con-
sideration of new kinds of model slightly reduces the threshold in the fit-ordering
that determines the class of interpretations such that an object must satisfy F un-
der each member of that class in order to count as clearly. . . clearly F. The clearly
clear Fs satisfy F under a more inclusive class of interpretations that do the (mere)
clear Fs; and the clearly clearly clear Fs satisfy F under a more inclusive class of
interpretations still. In each case, the more inclusive class contains those interpre-
tations that fit only slightly less well than the interpretations in the less inclusive
class. Limits on metasemantic gradualness will limit higher-order vagueness by
limiting how inclusive a class of interpretations can be obtained by successive iter-
ations of ‘clearly’.
2.9.9.2 Absolute clarity
This kind of view provides reason to deny the existence of ∆∗ or absolutely clear
Fs. Pre-fixing a sentence with another occurrence of ∆ requires truth under a more
inclusive classes of interpretations. The interpretations in successively more in-
clusive classes fit our use of language only slightly less well than do those in the
immediately preceding less inclusive class. If any interpretation is connected to
any other interpretation by a series of only slightly less well fitting interpretations,
then the absolutely clear Fs will be the objects that satisfy F under every interpre-
tation. Since nothing does so, there will be no absolutely clear of ∆∗ Fs. So even if
∆∗ and absolute clarity are precise, our actual vague classification won’t be.
Why think that the metasemantic facts are like this? Mightn’t some interpre-
Supervaluations 134
tations just be utterly irrelevant to our communication with the vague predicate
F? Maybe so. But that’s not what’s at issue. The question is whether the interpre-
tations that count x as an F are connected to those that don’t do so by a series of
successively less well-fitting interpretations, not whether such interpretations are
utterly irrelevant. If there’s no such series, then x will be absolutely clearly F. The
class of such objects will be the class of absolutely clear Fs, and there will be no
borderline cases to this class.
This kind of limit on higher-order vagueness differs from those argued for by
Williamson and Shapiro. Their arguments would make it logically impossible for
a language to lack sharp boundaries entirely. The present kind of restriction on
higher-order vagueness results from contingent features of the ordering of inter-
pretations by fit.
Does our use of ordinary vague predicates give rise to such disconnected series’
of interpretations? (And if it does, are the disconnected interpretations the only
ones that count certain objects one way or the other?) That’s what’s required for
the present kind of limit on higher-order vagueness. It’s hard to believe that our
use of language is like this. Small differences in the respects to which our use of F
is sensitive correlate with small differences in whether typical speakers would, by
and large, judge the objects in question to be F. And it is those judgements that
are primarily responsible for how well an interpretation fits our use of F. I can see
only one way of introducing a predicate that would generate discontinuities in the
fit-ordering of the kind necessary to restrict higher-order vagueness.
The method I have in mind introduces predicates by ostending a determinate
range of paradigm cases: this, that, the other and anything sufficiently similar to
them are all and only the Fs. Vagueness in ‘similar enough’ induces vagueness in
F. But any interpretation that places this, that and the other outside the extension
of F is utterly irrelevant to this use of F, and significantly less relevant than all
other interpretations. So this, that and the other are absolutely clearly F because no
Sorites series of gradually less well-fitting interpretations connects one that places
the paradigms outside the extension of F to one that places them inside it. Yet for
any other object, there may well be such a series. Hence this, that and the other will
be all and only the absolutely clear Fs. Here we have a sharp boundary resulting
Supervaluations 135
from our use of F.
This kind of case is atypical. Ordinary concepts are not introduced by osten-
sion of a determinate range of paradigms. As soon as a condition expressed by an
ordinary predicate is used in place of ostension when determining the paradigms,
an appropriate sorites series of interpretations will result, and so there will be no
absolutely clear cases. Our Lewisian Sharpening View thus combines with a reduc-
tive approach to metasemantics to generate a response to Williamson and Shapiro’s
arguments for hidden sharp boundaries in our actual vague classification.
2.9.10 Higher-order vagueness: concluding remarks
Does the Sharpening View or the Supertruth View provide the better approach to
higher-order vagueness? Well, the Supertruth View faces three problems.
Firstly, the Supertruth View’s account of higher-order vagueness undermines
its analysis of clarity in terms of truth and falsehood. Since all orders of borderline
case fall down a truth-value gap, there are distinctions marked by ∆ that cannot be
explained in terms of truth, falsity and gaps.
Secondly, the response to Shapiro’s argument requires supplementation with an
account of a non-Bivalent notion of truth (that supports LEM). §2.7.1 argued that
it is doubtful whether this is possible.
Thirdly, the response to Williamson’s argument for sharp boundaries requires
accepting a limit on how accurate the semantic theory can be. Although we shouldn’t
assume that a perfectly accurate mathematised semantic theory will be possible, a
view that purports to offer one is ceteris paribus preferable to one that does not.
None of these difficulties afflicts the Supertruth View. So that view is preferable.
2.10 Conclusion
We began with a formal setting and a range of philosophical interpretations that
might be imposed upon it. All bar two were ruled out in §§2.3–2.4. We then investi-
gated a range of difficulties for these two remainders. In each case, the Sharpening
View was seen to be less problematic than the Supertruth View. We also saw that
adequate responses to most problems with the Supertruth View require an account
Supervaluations 136
of its non-Bivalent conception of truth. It is not clear what this account might look
like.
The problems with explaining the Supertruth theorist’s conception of truth all
stem from the identification of clear truth with truth. This suggests that what the
Supertruth View really lacks is not so much an account of truth, but an account of
vagueness: the challenge of providing an account of clarity has simply been trans-
formed into the challenge of providing an account of truth. Hence the remainder
of this thesis will focus on the Sharpening View. We will, however, highlight those
points where the Supertruth View makes a difference (the next chapter’s discussion
of vague reference will contain quite a few).
One final question: to what extent is the Sharpening View a version of super-
valuationism? Keefe does not think it is. She calls it (or something very much
like it) the “pragmatic theory of vagueness” and opposes it to her own superval-
uationism (which is itself a version of our Supertruth View) (Keefe, 2000, ch.6).
The dispute is terminological. The views share (i) a formal structure, (ii) an anal-
ysis of clear truth as supertruth, and (iii) a non-privileging of any one sharpening
over any other in the apparatus of truth-evaluation. The differences concern only
the metaphysics of models and sharpenings, and whether the primary notion of
semantic evaluation is supertruth or s-truth. Furthermore, our Sharpening View,
unlike the Supertruth View does justice to the idea of vagueness as “semantic inde-
cision” or under-determination of content often associated with supervaluationism
(§2.3.3.3), and which Keefe herself endorses. It is therefore not misleading to de-
scribe the Sharpening View as a form of supervaluationism.
137
Chapter 3
Vagueness in Reference
This chapter and the next examine different applications of the Sharpening View
developed in the preceding chapter to the Problem of the Many. This chapter ex-
amines the idea that an ordinary object’s boundaries are vague insofar as it’s vague
which individual’s boundaries are at issue: Tibbles’s boundaries are vague because
it’s vague which object ‘Tibbles’ refers to. On this approach, Unger’s puzzle be-
comes a source of referential unclarity, and hence also of unclear boundaries. This
kind of view will ultimately be rejected.
This chapter also serves a second purpose: to defend and elaborate our Sharp-
ening View in response to several objections. These objections don’t concern the
Problem of the Many, so much as supervaluationist accounts of vague reference.
Since there may be sources of referential vagueness other than the Problem of the
Many, a full defence of the Sharpening View must address these objections.
Lewis endorses this reduction of vague boundaries to referential vagueness, and
describes two ways of applying it to the Problem of the Many. §3.1 begins by pre-
senting both and rejecting one. §3.2 then examines four objections to supervalua-
tionist accounts of referential vagueness. These concern: indirect reports of vague
speech; violation of a plausible constraint on reference; de re thought; conflict with
Direct Reference theory. Each objection will be found wanting. Three more seri-
ous difficulties for the Lewisian account of vague boundaries itself are presented
in §3.3. Doubts will be raised about whether this approach: provides a genuine
solution to the Problem of the Many, or merely makes it difficult to express it; can
Vague Reference 138
accommodate vagueness in the boundaries of self-referrers; is separable from ob-
jectionable components of Lewis’s metaphysical system. §3.4 concludes.
3.1 Two solutions
This chapter examines the following Proposal in the context of the Sharpening
View of vagueness that we developed in chapter 2:
Vagueness in the boundaries of ordinary objects results from vagueness about
which individual’s boundaries are in question,
Lewis (1993a) develops this Proposal in two ways. This section presents both and
rejects one.
3.1.1 Two options
This section introduces the Proposal in a little more detail.
Let h be one of Tibbles’s borderline hairs:
¬∆h is part of Tibbles ∧¬∆¬h is part of Tibbles.
It’s not plausible that vagueness in ‘h’ is responsible for this. So given the Sharpen-
ing View’s account of vagueness of as multiplicity of interpretation, there are two
options:1
(i) There are many intended interpretations of ‘Tibbles’.
(ii) There are many intended interpretations of ‘is part of’.
This chapter examines (i). The next chapter examines (ii).
Option (i) traces vagueness in Tibbles’s boundaries to vagueness about which
object ‘Tibbles’ refers to. This is Lewis’s view:
“The only intelligible account of vagueness locates it in our thought and
language. The reason it’s vague where the outback begins is not that
there’s this thing, the outback, with imprecise borders; rather, there are
1 Although these options could be combined, it’s (a) obscure why we would want to, and (b) clearer
to discuss them separately.
Vague Reference 139
many things, with different borders, and nobody has been fool enough
to try to enforce a choice of one of them as the official referent of the
word ‘outback’. Vagueness is semantic indecision.” (Lewis, 1986b, p.212)
Note that the argument from:
Vagueness is a feature of our thought and language.
and:
Vagueness is semantic indecision.
to:
Vague boundaries are the product of referential vagueness
is invalid without further premisses to rule out option (ii) above. One can con-
sistently endorse the first two theses despite rejecting the third. The next chapter
defends a view that does just that.
We focus on mereological vagueness, rather than vague boundaries or locations.
We lose no generality because (i) analogous remarks apply to these other kinds of
vagueness, and (ii) mereological vagueness implies vague boundaries and locations
via:
x is located in region r iff some part of x is located in region r.
So suppose h is Tibbles’s only borderline part. Let T+ be the fusion of h with the
rest of Tibbles. Let T− be T+ excluding h. Which of T+ and T− is Tibbles?2 And
which is a cat? Both are equally good candidates. The Lewisian Proposal is that
our use of ‘Tibbles’ doesn’t distinguish between them: each is an intended referent
of that name. Formally, this amounts to allowing supervaluationist models M that
violate the following constraint we imposed in §2.1:
For any sharpenings s, t ∈ M and singular term α : JαKs = JαKt.
Now, four desiderata:2 We needn’t assume that either T+ or T− is Tibbles. We could instead ask which constitutes
Tibbles, or take them as collections of microscopic particles and ask which composes Tibbles. The
formulation in the text simply aids presentation. See §1.1.2.2 for discussion.
Vague Reference 140
(D1) This should be borderline: ‘h is part of Tibbles’.
(D2) This should be clearly true: ‘Tibbles is a cat’.
(D3) This should be clearly true: ‘There is exactly one cat on the mat’.
(D4) This should be borderline: ‘h is part of the cat on the mat’.
Since h is part of T+ but not of T−, cross-sharpening variation about the referent
of ‘Tibbles’ secures (D1). Both approaches below agree on this. They diverge over
(D2)–(D4) and the treatment of ‘cat’. In the following, our quantifiers will often be
tacitly restricted to objects on the mat, thereby allowing us to drop the qualification
‘on the mat’: there is exactly one cat, and it’s borderline whether h is part of it.
Is there an argument for this Lewisian Proposal? One strategy compares it to
its rivals, arguing that, on balance, the Proposal is preferable. However, a more
satisfying approach would begin with an account of ordinary objects, or of material
reality more generally, that implies (or at least suggests) the Proposal; we would
like something more than an ad-hoc collection of theses unified only by their ability
to solve certain problems. I know of two closely related such accounts.
The first is Quine’s: a material objects is just “the material content of any por-
tion of space-time, however scattered and discontinuous” (Quine, 1976).3 Precision
in the boundaries of regions of space-time translates into precision in the bound-
aries of material objects. Vagueness in claims about the boundaries of those objects
must therefore come from vagueness as to their subjects.
The second account appeals to Lewisian views about the logicality and “onto-
logical innocence” of classical extensional mereology, and his use of that mereol-
ogy in the foundations of set-theory (Lewis, 1991, 1993b). Since logical and (pure)
mathematical vocabulary cannot, it seems, be vague, vagueness in mereological
predications must result from vagueness as to their subjects.
Neither Quine’s nor Lewis’s view carries intuitive force. Furthermore, the close
connections between the Proposal and other elements of Lewis’s metaphysical sys-
tem are already beginning to emerge. Our later arguments against the Proposal
(§3.3) therefore also count against these elements of Lewis’s system.4
3 A related view reduces objects to the regions of spacetime they occupy.4 §3.3.3 argues from the Proposal to counterpart-theory and perdurance. We then argue against
Vague Reference 141
3.1.2 Unger and Lewis
What is the relationship between Unger’s puzzle of too many candidates and Lewis’s
puzzle of vagueness and borderline candidates, given the Proposal? When there
are many best (and good enough) near-coincident cat-candidates on the mat, our
use of ‘Tibbles’ won’t distinguish amongst them. Each will be an intended refer-
ent of ‘Tibbles’. The result is referential unclarity in ‘Tibbles’ and unclarity about
Tibbles’s boundaries. Unger’s puzzle thus becomes a source referential and mereo-
logical unclarity.
It doesn’t follow that Unger’s puzzle is a source of referential vagueness be-
cause it doesn’t follow that ‘is part of Tibbles’ is Sorites-susceptible or admits of
higher-order borderline cases. That seems to require the gradualness of boundary-
transition that motivates Lewis’s puzzle (§1.2). But the end result is the same: our
use of ‘Tibbles’ isn’t fine-grained enough to distinguish one from amongst a range
of candidates on the mat. Since each candidate fits our use equally well (and well
enough), each is an intended referent for ‘Tibbles’ and contributes to unclarity in
mereological sentences featuring that name.
Unger argues that his and Lewis’s puzzles are distinct because his arises even
under the supposition that Tibbles’s boundaries are entirely precise (§1.3). On the
present approach, his reasoning is flawed. If Tibbles’s boundaries are precise, then
there’s a unique intended referent for ‘Tibbles’, and Unger’s puzzle doesn’t arise.
For if ‘Tibbles’ has a unique intended referent, then there’s a unique most cat-like
object on the mat. Were there several such objects, our use of ‘Tibbles’ wouldn’t
distinguish between them; so ‘Tibbles’ wouldn’t have a unique intended referent;
so Tibbles’s boundaries wouldn’t be precise. It is therefore safe to focus on Lewis’s
puzzle in the remainder.
3.1.3 One Cat
This section presents Lewis’s first account of mereological vagueness and the Prob-
lem of the Many. It combines two theses. The first concerns the extension of ‘cat’:
the Proposal by arguing against these views. Arguments against counterpart-theory and perdurance
thus translate into arguments against Quine’s and Lewis’s views about objects and mereology.
Vague Reference 142
Exactly one Tibbles-candidates satisfies ‘cat’ at each sharpening; different
candidates at different sharpenings (and each candidate at some sharpening).
Desideratum (D3) holds because this makes it supertrue that there’s exactly one cat
on the mat. Note however, that there’s nothing of which it’s supertrue that it is a
cat. Desideratum (D4) holds because ‘h is part of the cat’ is s-true iff the s-satisfier
of ‘cat’ includes h, and not all candidates do so.
The second thesis posits a penumbral connection between ‘cat’ and ‘Tibbles’:
For each sharpening s : J‘Tibbles’Ks ∈ J‘cat’Ks .
Desideratum (D2) holds because this makes it supertrue that Tibbles is a cat. This
constraint transfers referential vagueness in ‘Tibbles’, and hence mereological vague-
ness in Tibbles, to predicative vagueness in ‘cat’. This is as it should be: ‘Tibbles’
was introduced as a name for an individual cat.
We will call this the One Cat (OC) solution. On this approach, vague boundaries
reflect referential vagueness in names for ordinary objects. This referential vague-
ness induces corresponding predicative vagueness in our ordinary sortal concepts.
3.1.4 Many Cats
This section presents Lewis’s second account of mereological vagueness and the
Problem of the Many. The underlying idea is that every sufficiently cat-like object
counts as a cat. The Tibbles-candidates are all sufficiently cat-like. So they are all
cats: each satisfies ‘cat’ under each sharpening. Hence, for each sharpening s, the
s-referent of ‘Tibbles’ belongs to the extension of ‘cat’. So it’s supertrue, and hence
clearly true, that Tibbles is a cat. So desideratum (D2) is satisfied.
Let ‘Cx’ formalise ‘x is a cat’. Then the following is supertrue when restricted
to objects on the mat:
(1) ∃x∃y(Cx ∧ Cy ∧ x 6= y)
But recall desideratum (D3): it should be clearly true that there’s exactly one cat.
So Lewis must deny that (1) expresses the truth-condition of the English ‘there are
(at least) two cats’. An alternative truth-condition is required.
Vague Reference 143
Let ‘nco(x, y)’ formalise ‘x and y nearly materially coincide with one another’.
Lewis (1993a, p.178) suggests the following truth-condition for English numerical
claims ‘there are(at least) n F’s’:
∃x1 . . . ∃xn(Fx1 ∧ . . . ∧ Fxn ∧ ¬nco(x1, x2) ∧ ¬nco(x1, x3) ∧ . . . ∧ ¬nco(xn−1, xn))
English individuative vocabulary thus gets interpreted using near-coincidence rather
than numerical identity. Since T+ and T− are the only Tibbles-candidates, this is
supertrue:
∃x∀y(Cx ∧ (Cy→ nco(x, y))
So ‘there is exactly one cat’ is supertrue, despite both T+ and T− satisfying ‘cat’.
Hence (D3) is satisfied. Note also that, unlike the OC approach, it’s supertrue of
each cat-candidate that it’s a cat.
Desideratum (D4) is trickier. This should be borderline:
h is part of the cat.
Let ‘x ≤ y’ formalise ‘x is a (proper or improper) part of y’. Then we have two
Russellian truth-conditions:
∃x∀y(Cx ∧ [Cy→ y = x] ∧ h ≤ x)
∃x∀y(Cx ∧ [Cy→ (y = x ∧ h ≤ y)])
Although equivalent and supertruth-valueless on the OC approach, the present
approach makes them non-equivalent:
∃x∀y(Cx ∧ [Cy→ nco(y, x)] ∧ h ≤ x)
∃x∀y(Cx ∧ [Cy→ (nco(y, x) ∧ h ≤ y)])
The first is supertrue because every candidate/cat nearly coincides with a cat of
which h is part, namely T+. The latter is superfalse because h is not part of ev-
ery candidate/cat; specifically, it is not part of T−. Since it should be borderline
whether h is part of the cat, neither truth-condition is correct.
Lewis suggests two ways around this problem. The first treats descriptions as
singular terms, rather than disguised quantifier phrases: vagueness in ‘h is part of
Vague Reference 144
the cat’ is just like vagueness in ‘h is part of Tibbles’. Lewis’s second suggestion
is that both formalisations express intended interpretations of ‘h is part of the cat’:
our use of definite descriptions doesn’t distinguish between these truth-conditions.
This brings cross-sharpening variation in truth-value, thereby making it borderline
whether h is part of the cat. We should however, be sceptical of this second solu-
tion if a quantificational treatment of descriptions is attractive. The reason is that it
prevents a unitary analysis of vague boundaries: vagueness about Tibbles’s bound-
aries results from referential vagueness in ‘Tibbles’, while vagueness about the cat’s
boundaries results from vagueness in the truth-conditions of descriptions.
Like the OC approach, this Many Cat (MC) solution sees vague boundaries as a
reflection of referential vagueness (modulo the worries at the end of the preceding
paragraph). Unlike the OC approach however, this referential vagueness doesn’t
bring predicative vagueness in ordinary sortals, but a non-standard interpretation
of individuative vocabulary.
So, Lewis offers two ways to maintain that Tibbles is the only cat on the mat, de-
spite his vague boundaries. Both postulate referential vagueness in ‘Tibbles’. And
both employ supervaluations to make it borderline whether h is part of Tibbles.
They differ in two ways. Firstly, over the interpretation of ‘cat’: does it s-apply only
to the s-referent of ‘Tibbles’, or to every sufficiently cat-like object on the mat? Sec-
ondly, over the interpretation of English individuative vocabulary: does it express
identity and distinctness or near-coincidence and (extensive) disjointness?
Which approach is preferable? Lewis endorses both, arguing that different con-
texts require different solutions. The next section argues that he is wrong, and the
MC approach should be rejected.5
5 Williams (2006) offers a positive argument for the MC approach. His argument relies on at-
tributing the following conjecture to supervaluationists: ordinary speakers reason as if (clearly) true
existentials require (clearly) true instantiations. He claims that this is required by the best super-
valuationist response to the Sorites. But our Sharpening-theoretic response to the Sorites in §2.5.2
rejected that thesis in favour of an alternative. Williams’s argument for the MC approach is therefore
without force in the present context.
Vague Reference 145
3.1.5 The Problem of the Two
We’ve got two approaches to the Problem of the Many in place. Isn’t this one too
many? Lewis thinks not. He claims that different contexts require different so-
lutions, and hence that it’s context-sensitive which solution applies. This section
argues that he is wrong.
Lewis (1993a, p.180) claims that the MC approach is required when we dis-
cuss vagueness because attending to the equally cat-like natures of the candidates
places them all in the extension of ‘cat’. This can’t be right; for Lewis also dis-
cusses a case that the MC approach cannot accommodate, regardless of whether
we’re discussing vagueness. It follows that even if the choice between OC and MC
solutions is context-sensitive, it’s not sensitive to the difference between contexts
in which we’re discussing vagueness and more typical contexts (outside the phi-
losophy seminar): the MC approach cannot apply in every member of either class
of contexts. No such problem afflicts the OC approach. So unless it is context-
sensitive which solution applies, the MC approach fails and only the OC approach
is defensible. The defender of the MC approach therefore requires an alternative
account of when it applies. None is forthcoming.
If this right, then Lewis’s postulated context-sensitivity brings two problems.
Firstly, it multiplies senses of common nouns and individuative vocabulary with-
out necessity. Secondly, it undermines our semantic theory’s systematicity by posit-
ing context-sensitivity without an account of which features of context the sensi-
tivity is to, or why. Since the MC solution is defensible only by appeal to context-
sensitivity, it ought therefore to be rejected.
Here’s how Lewis presents the problematic case:
“Fred’s house taken as including the garage, and taken as not includ-
ing the garage, have equal claim to be his house. The claim had better
be good enough, else he has no house. So Fred has two houses. No!”
(Lewis, 1993a, pp.180–1)
Since the candidates don’t nearly coincide, the MC approach makes ‘Fred has two
houses’ supertrue. But whether we’re admiring Fred’s garden or discussing the se-
mantics of vagueness, that sentence ought to be false. So the MC approach doesn’t
Vague Reference 146
apply to all typical contexts, and it doesn’t apply whenever we’re discussing vague-
ness.6 So when does it apply? Without a well-motivated answer to this question
we ought to reject the MC approach. So let us do so.
3.2 Four problems with vague reference
This section rebuts four objections to supervaluationist accounts of vague refer-
ence, and hence also (indirectly) to the Lewisian Proposal about the Problem of the
Many. The objections concern: indirect speech reports in §3.2.1; a plausible con-
straint on reference in §3.2.2; de re thought in §3.2.3; Direct Reference in §3.2.4.
§3.3 turns to three more serious worries for the Lewisian approach to the Problem
of the Many itself.
3.2.1 Schiffer on speech reports
Stephen Schiffer (1998, §1; 2000, pp.321–6) argues that supervaluationism makes
indirect reports of vague speech false. §3.2.1.1 presents three problem cases.
§3.2.1.2 presents a simple semantics for indirect reports and a diagnosis of the
problem. §3.2.1.3 responds to Schiffer by implementing this semantics within our
Sharpening View. §3.2.1.4 addresses another difficulty Schiffer raises for this kind
of approach.
3.2.1.1 Three problem cases
This section presents three kinds of problem case. Here’s the first:
Pointing at a place, Al says to Bob: “Chris was there.”
Pointing at roughly the same place, Bob later reports this by saying: “Al said
that Chris was there.”6 Lewis (1993a, pp.179–80) claims that attending to vagueness places all the candidates in the
extension of the relevant sortal S. He then argues for the MC approach’s mereological interpretation
of individuation via the claim that such contexts require a sense in which there’s only one S. The
Problem of the Two undermines this argument: either no sense in which it’s true that there’s just one
S is required, or the mereological interpretation of individuation does not provide it.
Vague Reference 147
Schiffer claims the supervaluationist makes Bob’s report superfalse if ‘there’ is
vague. He argues as follows. Each sharpening assigns a different precisely de-
limited region of space to Bob’s utterance of ‘there’. But Al didn’t say of any precise
place that Chris was there. So each sharpening makes Bob’s report false. So Bob’s
report is superfalse.
That first case turns on referential vagueness concerning the demonstrative
‘there’. The problem generalises. Consider:
Anna says to Betty: “Chris is bald.”
Betty later reports this by saying: “Anna said that Chris is bald.”
Different sharpenings of Betty’s utterance of ‘bald’ assign it different precise ex-
tensions. But Anna didn’t say (anything to the effect) that Chris belongs to any
precise extension. So each sharpening makes Betty’s report false. So Betty’s report
is superfalse.
One final case:
Adama says to Bill: “Baldness is possessed by Chris.”
Bill later reports this by saying: “Adama said that baldness is possessed by
Chris.”
Different sharpenings of Bill’s utterance of ‘baldness’ assign it different precise
properties. But Adama didn’t say of any precise property that it is possessed by
Chris. So each sharpening makes Bill’s report false. So Bill’s report is superfalse.
Note the use of property-nominalisation to convert predicative vagueness into
referential vagueness. If Schiffer’s problem is genuine, then supervaluationism
faces a problem with referential vagueness, regardless of how it approaches the
Problem of the Many.
Although the first and third cases involve referential vagueness, the second
doesn’t. So why present this as a problem about referential vagueness? One an-
swer is that Schiffer’s (1998) paper denies the existence of vague places, despite
acknowledging vague properties and propositions. This creates a special problem
for the first case, not shared by the other two. Schiffer (2000) retracted this, but a
problem peculiar to vague names remains.
Vague Reference 148
Bob and Bill’s reports might be re-parsed thus:
“Al said, of that place there, that it’s where Chris was.”
“Adama said, of baldness, that it’s possessed by Chris.”
The truth of de re constructions like these is typically insensitive to which expres-
sions are used to denote the referents of the terms in the positions occupied by
‘that place there’ and ‘baldness’.7 So if these truths become false when terms like
‘precise place p’ and ‘precise property F’ are substituted into those positions, then
that must be because Al and Adama’s original statements weren’t about their refer-
ents. But since (i) each sharpening assigns a place or property to ‘that place there’
and ‘baldness’ which, let us imagine, is (or could in principle be) designated by
terms like those, and (ii) such substitutions do make Bob and Bill’s reports false, it
follows that (iii) each sharpening makes the reports false.
No similar re-parsing of Betty’s utterance to place ‘is bald’ in a position open
to substitution for co-designating (or even analytically coextensive) predicates is
possible. Changes in the truth-value of Betty’s report when ‘belongs to precise
extension e’ is substituted for ‘is bald’ therefore cannot be attributed to differences
in the semantic values of those expressions. We might therefore be unmoved by
the second case, despite finding the first and third persuasive.
3.2.1.2 Diagnosis
This section presents an account of the truth-conditions of indirect speech reports
and uses it to diagnose the source of Schiffer’s complaint.
What are the truth-conditions of indirect speech reports? Well, if Rosie said
that grass is green, then there is something Rosie said, namely, that grass is green.
This suggests that indirect reports ought to be construed as asserting the obtaining
of a relation between a speaker and a potential content, or proposition. The report
7 An example to illustrate. Suppose Lois Lane has never met Clark Kent, though she is an avid
follower of Superman’s adventures. Then if you and I know that Kent is Superman, we might report
her utterances to one another thus: “Lois said, of Kent, that he saved a child from a falling meteorite
yesterday.”
Vague Reference 149
will be true iff speaker and proposition really do stand in this relation. Let us
denote the proposition that p using ‘〈p〉’. Then the natural truth-condition is:8
pS said that Aq is true iff:
(i) S uttered a sentence that expressed 〈p〉; and
(ii) pthat Aq refers to 〈q〉; and
(iii) 〈p〉 = 〈q〉.
This isn’t uncontroversial, but it suffices for our purposes. It captures the idea
that the goal of an indirect report is to state the content of another’s utterance
(though not necessarily in the way that they did). An adequate semantics for indi-
rect reports must surely respect this. The apparatus of expressing and referring to
propositions merely provides a (natural and plausible) gloss on this.
This truth-condition needs supplementing with accounts of when a sentence
expresses a proposition, when a ‘that’-clause refers to a proposition, and when
〈p〉 = 〈q〉. The next section adds these to the Sharpening View. That framework
already provides a model of the association of linguistic items with contents. By
conceiving our formal object-language as used to make statements by a commu-
nity, these additions allow us to use it as a simple model of speech reports also. But
before this can provide a satisfactory response to Schiffer, we need to pinpoint the
source of his objection.
The problem is a mismatch between (a) the proposition expressed by the sen-
tence being reported, and (b) the propositions that sharpenings assign to the re-
port’s ‘that’-clause. Recall Schiffer’s claim that Al didn’t say of any precise place
that Chris was there, or equivalently, that Al’s utterance didn’t express any of the
precise singular propositions sharpenings assign to Bob’s utterance of ‘that Chris
was there’. Given the truth-condition above, this amounts to: Al uttered a sentence
that expressed 〈p〉, but there’s no sharpening s such that Bob’s ‘that’-clause s-refers
to 〈p〉. Why not? Schiffer seems to be assuming that Al’s utterance expressed a
single proposition (not about any precise region of space), while Bob’s ‘that’-clause
8 We make two simplifying assumptions. First assumption: sentences are uttered only in order
to make statements. Second assumption: sentences are used only literally. Our interest is in the
semantics of speech reports, not the analysis of various uses of language.
Vague Reference 150
vaguely refers to many precise propositions.9 Neither the Supertruth nor Sharpen-
ing theorist should grant this assumption.
The Supertruth theorist regards cross-sharpening variation about the referent
of a ‘that’-clause as representing vagueness about which proposition it picks out.
But on their view, each vague language has a unique intended vague semantic
structure: sentences aren’t vague because it’s vague which proposition they express,
but because they express vague propositions. The Supertruth theorist should there-
fore treat the referents of ‘that’-clauses as sharpening-invariant: for each sharpen-
ing s, pthat Aq s-refers to the vague proposition expressed by A. So no sharpening
assigns a precise proposition to Bob’s ‘that Chris was there’. And on no sharpening
does Bob report Al as saying something of any precise place. So Schiffer’s objection
fails.10 We won’t develop this further because we’ve rejected the Supertruth View.
So let us consider the Sharpening View instead.
3.2.1.3 Cure
We want to extend our Sharpening-theoretic model of vagueness to accommodate
indirect reports. If the truth-condition suggested above is correct, then it should
hold on every sharpening that respects the intended senses of ‘said’ and ‘that’-
clauses. Hence for any such sharpening s:11
9 Note that in order to state the problem, we’ve had to relativise ‘that’-clause reference to sharp-
enings without similarly relativising the expression of propositions by sentences, or which place Al
said something of. The responses canvassed here dissipate the problem by enforcing uniform de-
relativisation (on behalf of the Supertruth View) or uniform relativisation (on behalf of the Sharpen-
ing View).10 This might seem to introduce non-uniformity in truth-conditions: the s-referent of pthat Aq
isn’t the proposition s-expressed by A or determined solely by the s-values of the constituents of A.
A paratactic treatment of indirect reports avoids this. On this view, pS said that Aq decomposes
into two sentences. One is A. The other demonstrates A: S said thatRA. We can then regard
this demonstrative as indicating A in order to refer to the (vague) proposition it expresses, in much
the same way as we refer to colours by indicating objects that possess them. Since A itself isn’t
a constituent of the second sentence, there’s no argument for non-uniformity in truth-conditions.
Rumfitt (1993) develops a paratactic proposal along broadly similar lines.11 As stated, this truth-condition might be inadequate. It permits supertrue reports that are less
vague than the original statement: a report can be supertrue provided any subset of propositions
expressed by the original statement get assigned to its ‘that’-clause. Two responses are available. (a)
Vague Reference 151
pS said that Aq is s-true iff:
(i) S uttered a sentence B such that B s-expresses 〈p〉; and
(ii) pthat Aq s-refers to 〈q〉; and
(iii) 〈p〉 = 〈q〉.
Think of our formal object-language as used by a community. Then we can take
the uttering of sentences as part of our background metatheory, rather than as
analysable within the formal framework.12 Our framework already represents the
possible assignments of logically relevant content to linguistic items. We’ll add to
this an account of when a sentence expresses a proposition (under an assignment),
when a ‘that-’clause refers to a proposition (under an assignment), and also the
identity conditions of propositions. The result will hopefully be a reasonably ac-
curate representation of the semantics of speech reports that can accommodate the
clear truth of indirect reports of vague speech.
At the end of §3.2.1.2 we denied, on behalf of the Supertruth theorist, that
‘that’-clauses receive different propositions on different sharpenings. The present
response differs. We’ll vary the proposition expressed by the original sentence in
tandem with the proposition assigned to ‘that’-clauses used to report it.
When does a sentence express a proposition? Propositions are what speakers
say: if S said that A, then there’s something S said, namely 〈A〉. 〈A〉 is the con-
tent of A. Since the (logically relevant) content of a sentence is its truth-condition,
we can see proposition-talk as a nominalisation of truth-conditions. The Sharp-
ening View represents the association of sentences with truth-conditions using a
recursive definition of . So where ‘p’ stands in for a sentence in this definition:
Suppose the metasemantic facts are as M represents them. Then A s-expresses
〈p〉 iff: s, M A iff p.
Retain the truth-condition on the grounds that such reports are misleading, but not false. (b) Add
the following to (i)–(iii): if S uttered a sentence that, for some sharpening s, s-expresses 〈p〉, then, for
some sharpening t, pthat Aq t-refers to 〈p〉.12 For simplicity, we’ll homophonically translate names for speakers into the object-language and
ignore contextual variation in sentence-content between utterance and report.
Vague Reference 152
It doesn’t automatically follow that vague, and hence multiply interpreted, sen-
tences express many propositions because we haven’t yet given an account of when
〈p〉 = 〈q〉.
When does a ‘that’-clause refer to a proposition? The natural answer is that
they refer to the proposition expressed by the sentence from which they’re formed:
Suppose the metasemantic facts are as M represents them. Then pthat Aq
s-refers to 〈p〉 iff: s, M A iff p.
Putting these pieces together:
Suppose the metasemantic facts are as M represents them. Then pS said that
Aq is s-true iff:
(i) S uttered a sentence B such that: s, M B iff p; and
(ii) s, M A iff q, and
(iii) 〈p〉 = 〈q〉.
Two comments before we continue.
Firstly, this makes the reports in the original cases supertrue, even without an
account of proposition-identity. Those cases use the same sentence in the original
utterance and report.13 So the utterance s-expresses the same proposition as the
‘that’-clause s-refers to, for any sharpening s. An account of proposition-identity is
needed only when different sentences are used in the original statement and report
(since we’re ignoring context-sensitivity).
Secondly, consider a notion of proposition on which multiply interpreted sen-
tences do express many propositions. Our truth-condition makes it vague what
was said if the original statement was vague. For the extension of ‘said’, as defined
by clauses (i)–(iii), varies across sharpenings: if S said that A, then it’s s-true that S
said only what A s-expresses. We could avoid this by replacing (i) with:
(i′) S uttered a sentence B such that, for some sharpening t ∈ M : t, M A iff p.
13 The first case is tricky because it features demonstratives in the original utterance and report.
We’ll address that shortly.
Vague Reference 153
This brings the advantage of permitting a conception of speech-reports as an object-
language reflection of metalinguistic claims about intended interpretation.14 The
unmodified account cannot do so because claims about the range of intended inter-
pretations ought to be constant across those interpretations. Luckily, we need not
decide between (i) and (i′) here because both make the reports supertrue in each of
our three cases. So we’ll stick with the simpler condition (i).
Finally, to complete our response, we need an account of the identity conditions
of propositions. There are many options. Each seems to capture a legitimate notion
of content, or of what was said. Instead of arguing about the One True Notion of
Proposition, we can allow that different notions might be relevant to the various
projects in service of which different reports are made. (Moore, 1999, argues for
a similar view.) Different notions of proposition are characterisable using differ-
ent types of permissible transformation: 〈p〉 = 〈q〉 iff permissible transformations
convert p into q. Some candidate permissible transformations are:
The identity transformation: 〈p〉 = 〈q〉 iff p = q. This very fine-grained no-
tion is maximally sensitive to syntactic/compositional structure: 〈A ∨ B〉 6=
〈B ∨ A〉 when A 6= B.
As above, but also permutation of conjuncts and disjuncts. This notion is
less sensitive to compositional structure. Such propositions are roughly akin
to Fregean propositions in being individuated by presentations of semantic
values.15
As above, but also substitution of co-referential terms for elements of the do-
main. This roughly corresponds to a singular proposition in being insensitive
to how objects are designated.
As above, but also substitution of co-referential terms for set-theoretic con-
structs from the domain. This is akin to a Russellian proposition, a structured
complex of objects and properties.
14 The advantage is an explanation of a theoretical metasemantic concept in familiar terms.15 Variants allow substitution and insertion of double negation signs, and interchange of quanti-
fiers for their duals. We’ll assume the following notions permit these transformations.
Vague Reference 154
Interderivability. This corresponds to a conception of propositions as sets of
(logically) possible worlds.16
Now everything’s in place, let’s apply it to the first problem case in §3.2.1.1.
Brian Weatherson (2003, §1) suggests the following truth-condition:
Bob’s report ‘Al said that Chris was there’ is s-true iff: if Al’s utterance of
‘there’ s-refers to a place x, then so does Bob’s.
Think of ‘Chris was there’ as an atomic predication ‘F(there)’. Al’s statement and
Bob’s ‘that’-clause contain this same predicate, ‘Chris was (located at). . . ’. Do the
demonstratives in statement and report count as the same singular term? Let us
suppose not; maybe Al and Bob had to point in quite different directions. Then
Weatherson’s truth-condition for this particular case follows from our more gen-
eral account, provided that proposition-identity is insensitive to substitution of
co-referring terms for elements of the domain. Is Bob’s report supertrue? That
depends on whether the following penumbral connection holds:
For every sharpening s, Al’s ‘there’ s-refers to x iff Bob’s ‘there’ s-refers to x.
Since Bob pointed at roughly the same area as Al and intended to use his demon-
strative to report Al’s statement, it is very plausible that only interpretations that
respect this constraint will count as intended: Bob used his demonstrative defer-
entially to how Al used his. So Bob’s report is supertrue. Similar remarks apply to
the other two cases. So Schiffer’s objection fails.
3.2.1.4 Vague and precise contents
Our conclusion that Schiffer’s argument fails may have been too hasty. We assumed
that a report is s-true if s assigns the same proposition to its ‘that’-clause as to the
original statement. It’s unlikely this would satisfy Schiffer. He seems to deny that
any proposition assigned by any sharpening to a ‘that’-clause is also assigned to
a vague utterance: recall Schiffer’s claim that Bob didn’t say of any precise place
that Chris was there; his utterance didn’t express any precise proposition about
16 We ignore higher-order logics and incomplete deductive systems for simplicity.
Vague Reference 155
any precise place. But this is just the negation of the Sharpening View. It therefore
carries no weight without supporting argument.
An argument is nearby. Sharpenings assign precise places to Al’s ‘there’ and pre-
cise propositions to his ‘Chris was there’. This raises two worries. Firstly, it implies
that what Al said was precise, even though the sentence he used to say it was vague.
Secondly, how can Al’s statement be vague if what he said is precise? If these wor-
ries are genuine, then the Sharpening View’s account of vagueness collapses. This
section addresses this worry.
Note first that the Sharpening View analyses vagueness using a combination of
semantic and metasemantic concepts. The vague/precise classification therefore
primarily applies to content-bearers, not to their contents. We’ll use ‘presentation’
as a neutral term for any kind of content-bearer. Our first task is to extend the
vague/precise classification from presentations to contents. Three suggestions fol-
low.17
According to the first suggestion:
x is vague (precise) iff x is/could be the content of some vague (precise) pre-
sentation.
On this view, vagueness and precision aren’t mutually exclusive classifications of
contents.18 Since vague presentations always have vague contents, what Al said
was vague. But this doesn’t completely alleviate the problem because what Al said
may well also be precise.
According to the second suggestion:
x is vague (precise) iff x is/could be the content only of vague (precise) pre-
sentations (and is/could be the content of some vague (precise) presentation).
The closing parenthetical comment excludes trivially vague and precise contents.
On this view, vagueness and precision aren’t exhaustive classifications of contents.
17 A less concessive, though probably sound, response to Schiffer denies that there are any precise
contents on the grounds that only presentations can be vague or precise (and then, only in their role
as presentations). We won’t develop this here.18 Should vagueness and precision be mutually exclusive classifications of contents? That depends
on how deeply entrenched in our conceptual scheme that incompatibility is, as applied to contents
rather than presentations. I’m inclined to think that it’s not very deeply entrenched.
Vague Reference 156
Since Al uttered a vague sentence, what he said wasn’t precise. But there’s no guar-
antee that it will be vague either. So this suggestion doesn’t completely alleviate
the problem either.
A more promising suggestion modifies the logical form of attributions of vague-
ness to contents, by relativising them to presentations:
x is vague (precise) relative to α iff α is vague (precise) and x is a content of α.
On this view, a content can be vague relative to one presentation and precise rela-
tive to another. The important notion when assessing the vagueness of what some-
one said, is whether it was vague relative to the way they said it (whether they
conceptualised it as vague). Since Al’s ‘Chris was there’ is vague, so is what he said,
relative to the way he said it. And in this same sense, what Al said—i.e. that Chris
was there—isn’t precise either. On this approach, the propositions that sharpenings
assign to Al’s original statement and Bob’s report aren’t precise relative to either the
statement or report. The objection from the precision of those propositions and the
places they are about, therefore fails.
This provides the resources to respond to another of Schiffer’s objections. He
tries to commit the supervaluationist to truth-conditional ambiguity in the form
‘baldness is. . . ’. Consider:
Baldness is possessed by Chris.
Baldness is a vague property.
Suppose that ‘baldness’ is the only vague expression here. Then the first sentence
is clearly true iff each property each sharpening assigns to ‘baldness’ is possessed
by Chris. Since, we may suppose, Chris does possess each such property, the first is
clearly true. And the second sentence is clearly true iff each property each sharp-
ening assigns to ‘baldness’ is vague. But since, Schiffer claims, each such property
is precise, the second is clearly false. Since it should be clearly true, Schiffer claims
that its s-truth must turn not on whether the s-referent of ‘baldness’ belongs to the
s-extension of ‘is vague’, but on whether the word ‘baldness’ has many intended
interpretations.19 But then there is truth-conditional ambiguity in ‘baldness is. . . ’:
19 We treat ‘x is a vague property’ as analysable into ‘x is vague and x is a property’.
Vague Reference 157
the s-truth-conditions of only the first sentence displayed above turn on whether
the s-referent of ‘baldness’ belongs to the s-extension of the expression that re-
places the dots.
Schiffer’s argument is fallacious. He makes an unwarranted leap from the claim
that (a) the s-truth of ‘baldness is a vague property’ turns on whether ‘baldness’
has many intended interpretations, to the claim that (b) the s-truth-condition of
‘baldness is a vague property’ is not that the s-referent of ‘baldness’ belongs to the
s-extension of ‘is vague’. The Sharpening theorist grants (a) but may reject (b).
The s-extension of ‘is vague relative to’ is the class R of pairs 〈α, x〉 such that
α has many intended semantic values, one of which is x. We need to obtain an
s-extension for the unrelativised ‘is vague’ from this. Some means of closing the
second argument position of ‘is vague relative to’ is needed. The natural suggestion
is:
The s-extension of ‘is vague’, as that predicate occurs in pα is vagueq, is {x :
〈α, x〉 ∈ R}.
Since each property assigned by any sharpening to ‘baldness’ is vague relative to
‘baldness’, this makes it supertrue that baldness is a vague property. On this view,
the s-truth of ‘baldness is vague’ turns on whether ‘baldness’ has many intended
interpretations because the s-extension of ‘is vague’ does. But since the s-truth-
condition of ‘baldness is vague’ is that the s-referent of ‘baldness’ belongs to the
s-extension of ‘is vague’, there’s no argument for truth-conditional ambiguity here.
There is cross-sentence variation in the s-extension of ‘is vague’. This brings depar-
ture from compositionality: the s-truth-condition of pα is vagueq isn’t a function
of the s-values of α and ‘is vague’, but also of α itself. But since this brings no
loss of systematicity, there’s no reason to find it objectionable.20 Schiffer’s objection
therefore fails.20 Arguments from, e.g., the productivity of language to compositionality only seem to require
a finitely axiomatisable means of determining truth-conditions, not narrow compositionality of se-
mantic values.
Vague Reference 158
3.2.2 Barnett on incomplete definitions
David Barnett (2008) argues that supervaluationist accounts of vague reference are
incompatible with the following constraint on reference:
Referential Uniqueness (RU): A singular term refers only if features of its use
determine, of some unique thing, that that thing is its referent.
The incompatibility is supposed to arise because our use of ‘Tibbles’ does not de-
termine, of any Tibbles-candidate, that it is the referent of ‘Tibbles’; the candidates
are all on a par in that respect.
One line of response correlates (i) which object a sharpening s counts as being
determined as the referent of ‘Tibbles’ by our use of ‘Tibbles’, with (ii) the s-referent
of ‘Tibbles’:
For each term α and sharpening s, the s-extension of px determines that α
refers to yq is a relation that holds between our use of α and the s-referent of
α.
This makes the following supertrue: there is something x such that our use of
‘Tibbles’ determines that ‘Tibbles’ refers to x. But since ‘Tibbles’ refers to different
candidates on different sharpenings, there’s be no object x of which it’s supertrue
that our use of ‘Tibbles’ determines that ‘Tibbles’ refers to x. The de re formulation
of RU is intended to block this.
§1.4.7 presented a similar problem for singular thought, as opposed to linguis-
tic reference, taken from Unger (1980, §12A). Since no Tibbles-candidate is singled
out in preference to any other as the subject of our Tibbles-thoughts, Unger denies
that we have any singular thoughts about Tibbles. This problem is partly addressed
here, and partly in the next section.
Barnett also endorses a parallel constraint on predication that, by similar rea-
soning, should be incompatible with supervaluationism:
Predicative Uniqueness (PU): A predicate expresses a property only if features of
its use determine, of some unique property, that that property is expressed
by the predicate.
Vague Reference 159
PU isn’t needed to create trouble for the supervaluationist approach to predication
because a property-name formed by nominalising a vague predicate F will lack a
uniquely determined referent if F lacks a uniquely determined extension. Predica-
tive vagueness and property-nominalisation alone should suffice for conflict with
RU.
Barnett’s objection carries weight only against the Supertruth View; for only
on that view is no referent determined for a vague name. The next section elabo-
rates the Sharpening theorist’s response. We focus primarily on RU, though similar
remarks apply to PU and singular thought.
3.2.2.1 Solution
What is it for our use of α to determine, of x, that α refers to x? The answer is:
for some intended interpretation s, α s-refers to x. So the Sharpening View implies
that our use of ‘Tibbles’ does determine, of each Tibbles-candidate, that ‘Tibbles’
refers to it; for ‘Tibbles’ refers to each under some intended interpretation. Since
there’s no need to vary what counts as an intended interpretation across those in-
terpretations, it will even be supertrue of each candidate that our use of ‘Tibbles’
determines that ‘Tibbles’ refers to that candidate. Hence even the de re aspect of
RU is unproblematic.
In order to conflict with the Sharpening View, RU must therefore require that
use determine a unique intended interpretation. But then RU is just the negation of
the Sharpening View. Why should that be a constraint on reference? An argument
is required. Yet neither Barnett’s argument for RU nor the use to which he puts it,
requires uniqueness of intended interpretation. We begin with his uses of RU.
In §2 of his article, Barnett uses RU to argue that the following fail to introduce
referring names:
Let ‘Bitz’ name a resident of New York.
Let ‘Frib’ name a five-year-old child in Nigeria.
Let ‘Ball#1’ refer to one of the two balls in this urn.
Vague Reference 160
Barnett’s goal is to undermine arguments from incomplete stipulations to indeter-
minacy. His argument is, in essence, that these stipulations don’t determine any
intended interpretation because they fail to determine, of any object x, that the
term in question refers to x; for no object x do the stipulations make the state of
x relevant to the truth of sentences featuring ‘Bitz’, ‘Frib’ or ‘Ball#1’. A term that
lacks intended interpretation can hardly be a source of indeterminacy of meaning.
But it’s consistent with this that some uses of names might make more than one
interpretation intended, or more than one object relevant to the truth of a sentence
(featuring only that one name).
Here’s Barnett’s argument for RU:
“The constraint has the air of a truism. By definition a singular term
purports to refer to a single thing: if it has a referent, it has a unique
referent. And it is a platitude about meaning that words have their se-
mantic features determined solely by features of their use (where use is
construed broadly, to include both speaker intentions and relations to
their environment). Hence, if a singular term refers, features of its use
must determine a unique referent for it. (Do not confuse this constraint
with outright rejection of indeterminacy of reference; it does not ex-
clude indeterminacy as to which object is so uniquely determined.) We
have what appears to be a trivial constraint on reference for singular
terms.” (Barnett, 2008, p.173)
Construed as an argument for uniqueness of interpretation, this is fallacious.
Consider the first premiss: if a singular term has a referent, then it has a unique
referent. This has two disambiguations:
(i) If a singular term α has an intended interpretation, then α has a unique in-
tended interpretation, and that interpretation assigns α a unique referent.
(ii) If a singular term α has an intended interpretation, then that interpretation
assigns α a unique referent.
(i) is obviously question-begging in the present context. A persuasive argument can
therefore involve only (ii). But when combined with the claim that use alone de-
Vague Reference 161
termines meaning—that use alone determines intended interpretation—(ii) yields
only:
If a singular term α has an intended interpretation s, then use alone deter-
mines that s is an intended interpretation of α and s assigns α a unique refer-
ent.
This is compatible with a singular term having many intended interpretations. To
rule that out, Barnett needs the question-begging (i). He therefore provides no
argument for a reading of RU incompatible with the Sharpening theorist’s account
of vague reference.
3.2.3 McGee and McLaughlin on de re belief
De re and de dicto readings of ‘Ralph believes that Tibbles is a cat’ are typically
distinguished by quantifying into the scope of ‘believes’:
If Ralph believes that Tibbles is a cat, then Ralph’s belief is de re iff, for some
object x, Ralph believes that x is a cat.
McGee and McLaughlin (2000, pp.144-7) argue that this creates a problem for su-
pervaluationist accounts of vague reference.
3.2.3.1 The problem
Suppose Ralph believes that Tibbles is a cat. His belief is (super)true. The extension
of ‘cat’ varies across sharpenings. So there’s no object x of which it’s supertrue
that Ralph believes that x is a cat; for if there were, his belief wouldn’t be true on
all sharpenings. So Ralph’s belief is not de re. Since this turned on no features
specific to this case: de re belief about ordinary objects is impossible (or maybe just
extremely unlikely).
There is a subtlety here. Isn’t it true under each sharpening s that there’s some-
thing of which Ralph believes that it is a cat, namely the s-referent of ‘Tibbles’? If
so, then quantification-in is legitimate: it’s supertrue that, for some object x, Ralph
believes that x is a cat. Why isn’t this sufficient for Ralph’s belief to be de re?
Vague Reference 162
The reason is that McGee and McLaughlin deny that it’s true under any sharp-
ening that there’s something Ralph believes to be a cat:
“[T]here is no obvious way that supervaluation theory is going to help
us here. When we examine acceptable models [i.e. sharpenings], we
look at different ways of assigning sharp values to the terms of our lan-
guage. But assigning sharp values to the terms of our language doesn’t
do anything to sharpen the focus of Ralph’s beliefs. If A is an accept-
able model, A assigns a unique body of land to ‘Kiliminjaro’. But doing
this doesn’t do anything to answer the question whether Ralph’s belief
is about A(‘Kilimanjaro’)(+) or A(‘Kilimanjaro’)(-) [where these are two
nearly coincident Kilimanjaro-candidates].” (McGee and McLaughlin,
2000, p.146)
McGee and McLaughlin conceive sharpenings as formal representations of classical
semantic-structures that depart from the semantic properties of a vague language
only by settling all borderline cases. They don’t conceive supervaluationism as a
theory of vagueness, so much as a formal structure that resembles the structure of
vague thought and language in various respects. No sharpening makes it true that
there’s something Ralph believes to be a cat because sharpenings don’t modify, and
aren’t constituents of, the content of Ralph’s beliefs.
McGee and McLaughlin’s challenge thus concerns the object-directedness of de
re belief. In classical semantics, this coincides with quantification-in. Not so if
truth is supertruth: it can be supertrue that something is such that. . . , without
it being supertrue of anything that it. . . . Object-directedness isn’t guaranteed by
quantification-in alone, but by combining quantification-in with (super)truth:
If Ralph believes that Tibbles is a cat, then Ralph’s belief is object-directed iff
it’s (super)true of some object x that Ralph believes that x is a cat.
So on the Supertruth View, Ralph’s belief isn’t object-directed. But is it de re?
A positive answer requires an account of de re belief that doesn’t imply object-
directedness. We’ll leave that to the Supertruth theorists (Weatherson, 2003, §3
makes a start), and turn to the Sharpening View instead.
Vague Reference 163
3.2.3.2 Multiply interpreted de re belief
On the Sharpening View, vague content-bearers possess many contents. In par-
ticular, vague de re beliefs possess many singular contents. When Ralph believes
that Tibbles is a cat, he believes each singular classical proposition assigned by any
sharpening to ‘Tibbles is a cat’. The Sharpening theorist thus claims that Ralph has
many similar de re beliefs about the Tibbles-candidates.
Two comments. First, sharpenings must assign content to mental states as well
as linguistic items: vague thoughts and sentences both express many propositions.
Second, each of these singular propositions is object-directed. Ralph’s belief there-
fore permits quantification-in: for many objects x, Ralph believes that x is a cat.
McGee and McLaughlin disagree:
“The possibility that Ralph believes all of the countless singular propo-
sitions obtained by supplying Kilimanjaro candidates as arguments of
the proposition function [x is a mountain] can be readily dismissed, for
it implies that, no matter how careful and knowledgeable geographer
Ralph may be, his every true belief is accompanied by countless billions
of false beliefs.” (McGee and McLaughlin, 2000, p.146)
The problem is as follows. McGee and McLaughlin conceive the content of Ralph’s
cat-beliefs as given by a function f from objects x onto singular propositions such
that: f (x) is true iff x is a cat. Let T1, . . . , Tn be the Tibbles-candidates. Since there’s
only one cat on the mat, at most one proposition f (Ti) is true. So, contrary to our
claim above, Ralph doesn’t believe of each candidate that it is a cat; for if he did,
he would believe many false propositions.
The response is that our characterisation of the Sharpening theorist’s position
above was slightly misleading. On that view, the content of Ralph’s singular cat-
beliefs isn’t given by a single function from objects onto singular propositions, but
by a collection of functions f1, . . . , fn. Each fi maps each Tibbles-candidate Tj onto
a singular proposition fi(Tj). For each fi, exactly one of these propositions is true.
And for each Tibbles-candidate Tj, some proposition fi(Tj) is true. These functions
are the sharpenings of ‘cat’. The Sharpening theorist claims that Ralph believes
each true singular proposition obtained by supplying a Tibbles-candidate to one
Vague Reference 164
of these proposition functions. By sharpening the proposition function assigned to
‘cat’ and the candidate assigned to ‘Tibbles’ in tandem, Ralph’s belief can both have
many contents and be supertrue. Ralph doesn’t believe that many candidates are
cats, but his belief that Tibbles is a cat has many contents, each involving a slightly
different way some candidate is believed to be.
McGee and McLaughlin won’t permit this. They think of the content of Ralph’s
cat-beliefs as given by a single proposition function onto vague propositions. But
this aspect of the Supertruth View is just what the Sharpening theorist denies. The
Sharpening View faces no problem with de re belief.
3.2.4 Sorensen on Direct Reference
Roy Sorensen (2000) argues that supervaluationist accounts of vague reference are
incompatible with:
Direct Reference (DR): The semantic value of a name is its referent; names con-
tribute only their referents to the truth-conditions of sentences in which they
occur.
He describes a scenario in which some explorers introduce ‘Acme’ to name the first
tributary of the river Enigma, which they are about to begin charting.
“When [explorers] first travel up the river Enigma they finally reach the
first pair of river branches. They name one branch ‘Sumo’ and the other
‘Wilt’. Sumo is shorter but more voluminous than Wilt. This makes
Sumo and Wilt borderline cases of ‘tributary’.. . . ‘Acme’ definitely refers
to something, even though it is vague whether it refers to Sumo and
vague whether it refers to Wilt.” (Sorensen, 2000, p.180)
Assume DR: the semantic value of a name is its referent. Then each of ‘Acme’ and
‘Wilt’ contributes an object to the truth-conditions of ‘Acme is Wilt’. Since the ‘is’
of identity isn’t vague, ‘Acme is Wilt’ expresses a proposition of either the form
〈x = x〉 or the form 〈y = x〉.21 But such propositions cannot be sharpened. So the
supervaluationist technique cannot apply.21 Here we adopt Weatherson’s (2003) presentation for simplicity. Sorensen’s uses the idea that,
given DR, names function semantically like variables under an assignment.
Vague Reference 165
The example is poorly chosen. It is questionable whether ‘Acme’ is semantically
directly referential, as opposed to a rigidified description. But we can set this aside;
for if any names are directly referential, then names for ordinary objects surely
are; in which case, the Lewisian Proposal makes ‘Tibbles’ relevantly analogous to
Sorensen’s treatment of ‘Acme’.
Only the Supertruth theorist should be moved by this, though not as it stands.
They should reject the (classical) conception of propositions Sorensen assumes.
Propositions should instead be understood in terms of the supertruth- and superfalsity-
conditions assigned them by supervaluationist models. On this approach, an ex-
pression’s semantic contribution is its role in delimiting the space of sharpenings.
How does a name contribute to this? A natural answer is: a name contributes a class
of objects, its candidate referents. But this isn’t quite right because ‘Tibbles’ and
‘cat’ are penumbrally connected. On the Supertruth View, ‘Tibbles’ makes at least
two semantic contributions: (i) its candidate referents; (ii) its penumbral connec-
tions to other expressions. Since penumbral connections are analytic connections
between meanings, contribution (ii) is incompatible with DR.
On the Sharpening View, DR ought to be relativised to intended interpretations:
Relativised Direct Reference (RDR): For any intended interpretation s, a name’s
sole contribution to s-truth-conditions is its s-referent.
On one sharpening, ‘Acme’ contributes Sumo to the truth-conditions of ‘Acme is
Wilt’. On another, it contributes Wilt. On the first sharpening, ‘Acme is Wilt’
expresses a proposition of the form 〈y = x〉. On the second, it expresses a proposi-
tion of the form 〈x = x〉. ‘Acme’ complies with RDR on both sharpenings, though
‘Acme is Wilt’ has a different truth-value on each. So RDR is compatible with ‘Acme
is Wilt’ being borderline. The key point is that sentences get supervalued, not the
propositions they express. And since penumbral connections are constraints on
interpretations, not features of them, they present no threat to RDR.
Weatherson introduces a variant puzzle:
“[O]ne quite plausible principle about precisifications is that precisifi-
cations must not change the meaning of a term: they may merely pro-
vide referents where none exists. Now the supervaluationist has a prob-
Vague Reference 166
lem. For it is true that one of [‘Acme is Wilt’ and ‘Acme is Sumo’] is true
in virtue of its meaning, since its meaning determines that it expresses a
proposition of the form 〈x = x〉. But each sentence is false on some pre-
cisifications, so some precisifications change the meanings of the terms
involved.. . . The best way to respond to this objection is simply to bite
the bullet.” (Weatherson, 2003, p.498)
The problem is illusory. The Supertruth theorist should deny that the meaning of
either ‘Acme is Wilt’ or ‘Acme is Sumo’ is a proposition of the form 〈x = x〉. Hence
neither need be true in virtue of its meaning. And the Sharpening theorist should
deny that sharpenings provide referents where none exists. Although different
sharpenings assign different referents, they don’t “fill in the gaps” left by some
other reference relation that isn’t relativised to sharpenings: s-reference is the only
semantic notion of reference. So no bullet-biting is required.
3.3 Three problems with the Problem of the Many
We’ve seen four objections to supervaluationist accounts of vague reference. The
Sharpening theorist should not find them compelling. This leaves us clear to fol-
low the Lewisian Proposal and employ referential vagueness in an analysis of vague
boundaries and response to the Problem of the Many. This section presents three
problems for this Proposal. §3.3.1 questions whether it provides a genuine solu-
tion to the problem. §3.3.2 argues that it cannot accommodate vagueness in the
boundaries of self-referrers. And §3.3.3 argues that it brings objectionable com-
mitments in the metaphysics and semantics of time and modality. We ought to
seek an alternative.
3.3.1 A genuine solution?
Is the Lewisian Proposal a genuine solution to the Problem of the Many? There
are good, but inconclusive, reasons to think not. This section approaches them
via a dilemma due to Neil McKinnon (2002). §3.3.1.1 presents the first horn, on
which sharpenings are principled: objects that satisfy the same predicate share some
(not overly disjunctive) property that suffices for satisfying that predicate. §3.3.1.2
Vague Reference 167
presents the second horn: sharpenings are unprincipled. McKinnon argues that
neither horn is acceptable. Even if he is wrong, only the second horn is tenable.
§3.3.1.3 shows that this commits the supervaluationist to the extrinsicality of the
property(s) expressed by ‘cat’. §3.3.1.4 suggests that this undermines the claim that
the present approach provides a genuine solution.
3.3.1.1 McKinnon’s first horn: principled sharpenings
Suppose there are two cats on the mat, Tibbles and Sophie. The following should
be clearly true:
There are exactly two cats on the mat.
So for each sharpening s, exactly one Tibbles-candidate and one Sophie-candidate
s-satisfy ‘cat’. Suppose that sharpenings are principled. Then if x and y both s-
satisfy ‘cat’, then they share some property sufficient for s-satisfaction of ‘cat’. But
any relevant property shared by a Tibbles-candidate and a Sophie-candidate will
also be shared by all the Tibbles-candidates. For the Tibbles-candidates are much
more like one another in cat-respects than they are any Sophie-candidate. (Perhaps
Tibbles and Sophie belong to different breeds.) So all the Tibbles-candidates s-
satisfy ‘cat’. Since s was arbitrary: it’s superfalse that there are exactly two cats on
the mat.
The defender of principled sharpenings has two options. The first is simply
to hope that there are suitable properties to provide enough principled sharpen-
ings. Such wishful thinking should be given no credence. The second is to present
enough examples to make it plausible that there are sufficient properties to provide
enough principled sharpenings to accommodate all of Tibbles and Sophie’s evident
vagueness. McKinnon considers and rejects several candidates, e.g.: specific shapes
and ratios of interior to exterior densities. The candidates seem either so specific as
to prevent co-satisfaction of ‘cat’ by a Tibbles-candidate and a Sophie-candidate, or
there’s no guarantee that they’ll be possessed by exactly one candidate from each
collection. So let us set principled sharpenings aside.
Vague Reference 168
3.3.1.2 McKinnon’s second horn: arbitrary sharpenings
McKinnon (2002, §3) endorses:
Non-arbitrary differences (NAD) For any cat and non-cat, there is a principled
difference between them in virtue of which the one is a cat and the other not.
Non-arbitrary similarities (NAS) For any two cats, there is a principled similarity
between them in virtue of which they are both cats.
These induce penumbral connections between ‘Tibbles’ and ‘Sophie’ that McKin-
non thinks spell trouble for the present approach.
Suppose that Tibbles-candidate T+ s-satisfies ‘cat’, while Tibbles-candidate T−
does not. Then by NAD: there is a principled difference between T+ and T− in
virtue of which this is so. But T+ resembles T− much more closely in cat-respects
than either does any Sophie-candidate. So each principled difference between T+
and T− is also a principled difference between T+ and each Sophie-candidate. So
no Sophie-candidate s-satisfies ‘cat’.22 Since s was arbitrary: it’s superfalse that
there are exactly two cats on the mat.
Now suppose that Tibbles-candidate T+ and Sophie-candidate S both s-satisfy
‘cat’. Then by NAS: there is a principled similarity between T+ and S in virtue of
which this is so. But T+ resembles T− much more closely in cat-respects than it
does any Sophie-candidate. So each principled similarity between T+ and S is also
a principled similarity between T+ and T−. So T− also s-satisfies ‘cat’. Since s was
arbitrary: it’s superfalse that there are exactly two cats on the mat.
3.3.1.3 Extensive overlap and intrinsicality
Weatherson (2003, §5) responds to the second horn by, in effect, varying what
counts as a principled difference across sharpenings. He begins with the equiv-
alence:
x is a cat iff x is a cat-candidate that does not extensively overlap any cat
(other than x).
22 This assumes that if something is an F in virtue of being a G, then being a G suffices for being
an F.
Vague Reference 169
This is true on all sharpenings that place exactly one Tibbles-candidate and one
Sophie-candidate into the extension of ‘cat’. It’s true on any such sharpening that
any two cats resemble one another in respect of their not extensively overlapping
any cat other than themselves. Such sharpenings also make it true that any cat
and any cat-candidate that fails to be a cat differ in that same respect. Hence,
Weatherson claims, NAD and NAS are supertrue. There are two problems.
The first problem is that NAD and NAS don’t involve material biconditionals,
but an ‘in virtue of’ locution. They don’t simply assert the existence of principled
similarities and differences, but claim that similarities and differences in respect
of being cats obtain in virtue of these other principled similarities and differences.
But the Problem of the Many arises because the condition x is a cat-candidate that
does not overlap any cat (other than itself) determines a unique Tibbles-candidate
only given a prior selection of some unique candidate as a cat. So on Weatherson’s
proposal, similarities and differences in respect of being a cat-candidate that does
not extensively overlap any cat (other than itself) obtain in virtue of similarities
and differences in respect of being a cat, rather than, as NAD and NAS require, vice
versa. Weatherson has two responses available.
Firstly, he may point out that ‘in virtue of’ locutions are notoriously murky. It’s
far from obvious what their content amounts to, or what constrains their correct
usage. If these doubts are well-founded, then only the weaker biconditional read-
ings of NAD and NAS may legitimately be insisted on. Secondly, he may (and does)
claim that the stronger reading amounts to a demand for a (possibly partial) anal-
ysis of ‘cat’. There is no reason to expect that this will be possible, especially for
an apparent natural kind term like ‘cat’. So let us turn to the second problem for
Weatherson’s proposal.
Weatherson’s proposal brings cross-sharpening variation in the respects of re-
semblance and difference that satisfy NAD and NAS. On sharpening s, cats re-
semble one another in respect of not extensively overlapping any cat-candidate to
which ‘cat’ s-applies. And on sharpening t, cats resemble one another in respect of
not extensively overlapping any cat-candidate to which ‘cat’ t-applies. Since s and
t assign different extensions to ‘cat’, these are different respects of resemblance.
But such similarities and differences obtain only because of the existence of the
Vague Reference 170
semantic structures s and t. They are no more genuine, principled or intrinsic re-
semblances and differences than x and y’s resemblance in respect of belonging to
{x, y}, or their difference in respect of belonging to {x}. If ‘cat’ distinguishes be-
tween objects on such slim grounds, then it does not mark a genuine, principled or
intrinsic distinction. The only such distinction in the vicinity is that between the
cat-candidates and everything else. But that’s not marked by ‘cat’ on any sharpen-
ing.
So if sharpenings are unprincipled, then ‘cat’ does not mark a natural kind
or intrinsic property, regardless of whether McKinnon is right to insist on NAD
and NAS. Is this objectionable? One reason to think not appeals to the apparent
maximality of ‘cat’. But as §1.1.4.2 showed, the argument from maximality to non-
intrinsicality assumes that maximality is a semantic feature of predicates, rather
than a metaphysical feature of the boundaries of objects. This is tantamount to
assuming that the property expressed by ‘cat’ is extrinsic. Maximality therefore
does not provide independent or theory-neutral reason to deny that the cats form
a natural kind.
Regardless of whether we ought to deny that the cats form a natural kind, ad-
vocates of the present must do so. This raises serious doubts about whether it
provides a genuine solution to the Problem of the Many.
3.3.1.4 Genuine solution or semantic trickery?
§1.4 showed that the Problem of the Many generates conflict with:
There is not widespread causal overdetermination of the effects of ordinary
objects by ordinary objects.
Your actions are free, in the sense of not being entailed by those of any person
distinct from you.
You make real choices, independent of those made by any other conscious
being.
It is an undoubtable Moorean fact that I’m the only person in my chair.
Our ordinary moral judgements are not in radical error.
Vague Reference 171
You are the only thinking and experiencing conscious being in your chair.
Penumbral constraints like the following resolve this conflict by making the claims
above supertrue:
The s-extension of ‘ordinary object’ is the union of the s-extensions of all
ordinary sortals, ‘cat’, ‘human’, ‘dog’,. . . .
The s-referent of ‘you’ is the only one of your person-candidates in the s-
extensions of ‘person’, ‘conscious’, ‘experiencer’, ‘chooser’ and ‘thinker’.
The s-referent of ‘I’ is the only one of my person-candidates in the s-extension
of ‘person’.
The s-extension of, for example, ‘murderer’ is a subset of the s-extension of
‘person’.
Does this semantic technique solve the initial problems? That depends on what
those problems are.
On one account, the Problem of the Many is a problem because it seems to show
that certain sentences possess truth-values which they ought not to. If this is right,
then we have a genuine solution because the supervaluationist technique ensures
a proper distribution of truth-values (though §3.3.2 and the end of the present
section question even this). But this is not the only account.
On an alternative account, the Problem of the Many is problematic because it
implies an overabundance of certain kinds of object. The preceding section showed
that the supervaluationist technique does not address this. Although it’s not true
on the present approach that my person-candidates are all conscious, they are in-
trinsically just like conscious beings; let us say that they are conscious∗. If there’s
any important or intrinsic distinction amongst objects in the vicinity of conscious-
ness, it’s that between the conscious∗ things and the rest.23 Yet I’m not the only
conscious∗ being in my chair; my actions and choices are entailed by the actions∗
and choices∗ of many other conscious∗ beings. Is this problematic? If so, then the
23 An important distinction amongst objects contrasts with a distinction amongst objects in respect
of their relationships to potential semantic structures and our use of language.
Vague Reference 172
Lewisian Proposal merely disguises the problem, rather than addressing it. In ef-
fect, the ∗-ed properties provide a significant sense in which it’s supertrue that, say,
many cats are on Tibbles’s mat: there are many cat∗’s on Tibbles’s mat. It must be
shown that this is unproblematic before we can regard the Lewisian Proposal as a
genuine solution.
Now, ‘conscious∗’ is a theoretical term, not part of ordinary vocabulary. So
one might deny that we’re entitled to any intuitive judgements about whether our
near-coincidence with many conscious∗ beings is problematic. In which case, ∗-ed
analogues of the problems above carry no intuitive force. This may be right. But it
doesn’t show that those ∗-ed analogues aren’t problems. It shows at most that we
aren’t entitled to a view either way; in which case, we aren’t entitled to a view about
the success of the Lewisian Proposal either. Without a proper investigation of the
metaphysics of causation, action, free will, choice, consciousness, personhood and
morality, endorsing that Proposal amounts to simply closing one’s eyes and hoping
for the best. Furthermore, there is reason to think that we are entitled to just the
same intuitive judgements about ∗-ed concepts, as we are their ∗-less counterparts.
Are our beliefs about cats beliefs about a certain kind of object, or about the
truth-values of sentences? Surely not the latter. Yet the Lewisian Proposal implies
that ‘cat’ and ‘conscious’ don’t mark genuine kinds. Cat∗s and conscious∗ beings
are just like cats and conscious beings in every respect that’s relevant to the justi-
fication of our beliefs about cats and conscious beings. So the justification for our
beliefs about cats and conscious beings extends to justify the same beliefs about
cat∗s and conscious∗ beings; in which case, we are justified in regarding the ∗-ed
variants of the problems above as genuine problems, if we are justified in so regard-
ing the originals. If this is right, then the Lewisian Proposal doesn’t even assign the
right truth-values to sentences, never mind resolve any underlying metaphysical
problems.
Although this last argument certainly isn’t beyond reproach, a proper investi-
gation lies beyond the scope of this thesis. But it does suggest that the burden of
proof lies with the Lewisian to show that their “solution” is genuine. It also exposes
the Proposal’s hidden assumptions about an array of core philosophical disputes.
We ought not to endorse it without having undertaken a proper investigation of
Vague Reference 173
those issues.
3.3.2 Hawthorne on self-reference
A variant on an argument of John Hawthorne’s makes trouble for the One Cat so-
lution (Hawthorne, 2006a). Hawthorne’s argument targeted the following combi-
nation of views:
Vague languages satisfy classical logic and semantics, and have a unique in-
tended interpretation.
Vague expressions are semantically plastic, i.e. sensitive to indiscriminably
slight variations in use. (This was used to justify the idea that the languages
of different speakers might have slightly different intended interpretations.)
Vague boundaries involve vagueness about which object’s boundaries are at
issue.
Our variant differs from Hawthorne’s by (i) not requiring a unique intended inter-
pretation, (ii) not assuming semantic plasticity, and (iii) not assuming that the lan-
guages of different speakers can have slightly different intended interpretations.
Unlike Hawthorne’s argument, ours will apply to ourselves as well as to other
speakers.
§3.3.2.1 begins with three versions of the argument. §§3.3.2.2–3.3.2.5 consider
and reject four responses.
3.3.2.1 The argument
This section presents three versions of Hawthorne’s argument.
Version one: Paula Suppose that Paula is sitting on a chair. Suppose also that her
boundaries are vague such that there are two Paula-candidates, P+ and P−. Since
Paula is a person, and clearly the only person sitting on her chair, P+ and P− are
also the only person-candidates. So there are two sharpening s+, s− of our and
Paula’s shared vague language:
‘Paula’ s+-refers to P+.
Vague Reference 174
‘Person’ s+-applies to P+.
‘Paula’ s−-refers to P−.
‘Person’ s−-applies to P−.
Given this, the argument rest on four seemingly obvious truths:
(2) Paula is the only object in her chair that can speak and think.
(3) Utterances of ‘I’ by Paula refer to Paula.
(4) Utterances of the form ‘a is F’ by Paula are true iff ‘F’ applies to the referent
of ‘a’, as Paula uses those expressions.
(5) Our and Paula’s language is vague: s+ and s− are its intended interpretations.
Here’s how the argument goes.
Suppose it’s clearly true that Paula says “I am a person”. Each of (2)–(5) are
also clearly true. So each is true on each sharpening. We want to know: is it
clearly true that Paula’s utterance was clearly true? Is it true on each sharpening
that Paula’s utterance is true on each sharpening? By (5): s+ is a sharpening of our
language. So is it true on s+ that Paula’s utterance was true on each sharpening?
The following argument suggests that it isn’t, and hence that it isn’t clearly true
that Paula’s utterance was clearly true.
On s+, ‘Paula’ refers to P+. By (2): Paula/P+, and nobody else, said ‘I am a
person’. By (5): s− is a sharpening of Paula/P+’s language. By (4): Paula/P+’s
utterance of ‘I am a person’ is true on s− iff ‘person’ s−-applies to the referent of ‘I’,
as Paula/P+ uses ‘I’. By (3): ‘I’ refers to Paula/P+, as Paula/P+ uses ‘I’. But ‘person’
doesn’t s−-apply to P+. So it isn’t true on s+ that Paula/P+’s utterance of ‘I am a
person’ is true on s−. So it isn’t true on s+ that Paula’s utterance was supertrue.
So it isn’t clearly true that Paula’s utterance of ‘I am a person’ was clearly true. In
fact, it’s true on each sharpening that Paula’s utterance is true on some but not all
sharpenings. So it’s clearly true that Paula’s utterance was borderline. Yet surely it
shouldn’t be.
Hawthorne’s argument delivers this same conclusion. Unlike Hawthorne’s ar-
gument however, ours applies to ourselves as well as to Paula.
Vague Reference 175
Version two: me Suppose my boundaries are vague such that there are two Nick-
candidates N+, N− in my chair. Since I am a person and clearly the only person
in my chair, N+ and N− are also the only person-candidates. So there are two
sharpenings of ‘person’ in my language:
‘Person’ s+-applies to N+.
‘Person’ s−-applies to N−.
Five seemingly obvious truths:
(6) I am the person in my chair.
(7) Only the person in my chair can speak and think.
(8) Utterances of ‘I’ by the person in my chair refer to that person.
(9) Utterances of the form ‘a is F’ by the person in my chair are true iff ‘F’ applies
to the referent of ‘a’, as the person in my chair uses those expressions.
(10) My/the person in my chair’s language is vague: s+ and s− are its intended
interpretations.
Suppose it’s clearly true that I say “I am a person”. Each of (6)–(10) are clearly true.
So each is true on each sharpening. Is it clearly true that my utterance was clearly
true? Is it true on each sharpening of my language that my utterance was true on
each sharpening? By (10): s+ is a sharpening of my language. So is it true on s+
that my utterance was true on each sharpening? The following argument suggests
that it isn’t, and hence that it isn’t clearly true that my utterance was clearly true.
On s+, ‘person’ applies to N+. Hence by (6) and (7): the person in my chair/N+,
and nobody else, said ‘I am a person’.24 By (10): s− is a sharpening of the person
in my chair/N+’s language. By (9): the person in my chair/N+’s utterance of ‘I am
a person’ is true on s− iff ‘person’ s−-applies to the referent of ‘I’, as the person in
my chair/N+ uses ‘I’. By (8): ‘I’ refers to N+, as the person in my chair/N+ uses
‘I’. But ‘person’ doesn’t s−-apply to N+. So it isn’t true on s+ that the person in
24 The extra premiss (6) is needed to fix a referent for ‘I’. It ensures that its s+-referent is the unique
s+-satisfier of ‘person’, namely N+.
Vague Reference 176
my chair/N+/my utterance of ‘I am a person’ was true on s−. So it isn’t true on s+
that my utterance was supertrue. So it isn’t clearly true that my utterance of ‘I am
a person’ was clearly true. In fact, it’s true on each sharpening that my utterance of
‘I am a person’ is true on some but not all sharpenings. So it’s clearly true that my
utterance was borderline. Surely it shouldn’t be.
It might seem that this relies on ‘person’ applying to only one of my person-
candidates on each sharpening, and hence that the Many Cat approach avoids it.
The final variant shows that this is mistaken.
Version three: Nick Stick with the same Nick/person-candidates as before. Two
more facts about s+ and s−:
‘Nick’ s+-refers to N+.
‘Nick’ s−-refers to N−.
Five seemingly obvious truths:
(11) I am the person in my chair.
(12) Only the person in my chair can speak and think.
(13) Utterances of ‘I’ by the person in my chair refer to that person.
(14) Utterances of the form ‘a is b’ by the person in my chair are true iff the referent
of ‘a’ is identical to the referent of ‘b’, as the person in my chair uses those
expressions.
(15) My/the person in my chair’s language is vague: s+ and s− are its intended
interpretations.
Suppose it’s clearly true that I say “I am Nick”. This and each of (11)–(15) are
clearly true, hence true on each sharpening. Is it clearly true that my utterance was
clearly true? Is it true on each sharpening of my language that that my utterance
was true on each sharpening? The following argument suggests that it isn’t, and
hence that it isn’t clearly true that my utterance was clearly true.
On s+, ‘person’ applies only to N+. Hence by (11) and (12): the person in my
chair/N+, and nobody else, said ‘I am Nick’. By (15): s− is a sharpening of the
Vague Reference 177
person in my chair/N+’s language. By (14): the person in my chair/N+’s utterance
of ‘I am Nick’ is true on s− iff the s−-referent of ‘Nick’ is identical to the referent of
‘I’, as the person in my chair/N+ uses ‘I’.25 By (13): ‘I’ refers to N+, as the person
in my chair/N+ uses ‘I’. But ‘Nick’ doesn’t s−-refer to N+. So it isn’t true on s+ that
the person in my chair/N+’s utterance of ‘I am Nick’ was true on s−. So it isn’t true
on s+ that my utterance was supertrue. So it isn’t clearly true that my utterance of
‘I am Nick’ was clearly true. In fact, it’s true on each sharpening that my utterance
was true on some but not all sharpenings. So it’s clearly true that my utterance of
‘I am Nick’ was borderline. So who am I?
Since the Many Cat approach makes no difference to the interpretation of names,
it doesn’t undermine this version of the argument. One might respond by point-
ing out that (14) interprets individuative apparatus using identity, when the MC
approach interprets it as near-coincidence.
This response fails because there’s a variant of the Problem of the Two on which
my leg is being amputated (under powerful local anaesthetic). Halfway through the
operation, I say “I am Nick”. But my body taken as excluding my leg (N−) and my
body taken as including my leg (N+) aren’t nearly coincident. So the referent of ‘I’,
as the person in my chair/N+ uses ‘I’, isn’t nearly coincident with the s−-referent
of ‘Nick’. So it still isn’t true on s+ that my utterance of ‘I am Nick’ was true on s−.
What responses are available? Let’s focus on the first version for simplicity.
Since it rests on four premisses, there are four options. None is attractive.
3.3.2.2 First response: deny (2)
This response denies that Paula is the only object in her chair that can think and
speak. This alone won’t resolve the problem. We assumed that it’s clearly true that
Paula said “I am a person”. So the argument still shows that her utterance wasn’t
clearly true; but it also shows that several other speaker’s utterances of ‘I am a
person’ weren’t clearly true too. To succeed, this response must deny that it’s clearly
true that Paula spoke at all. This is a significant cost. The analogous response to
25 Although ‘I am Nick’ isn’t of the form ‘a is b’, the difference is merely typographical.
Vague Reference 178
the second and third versions of the problem implies that it’s not clearly true that
it was me who spoke. So who did?
Even this doesn’t solve the problem however. For whoever spoke, their lan-
guage is vague such that, on at least one intended interpretation of their language,
‘person’ doesn’t apply to them.26
Even if this response could make Paula’s utterance clearly true, there would
be sharpenings of her language on which it’s true that her actions are entailed by
those of someone else, specifically the person in her chair. So she wouldn’t clearly
act freely. In light of these problems, we should reject this response.
3.3.2.3 Second response: deny (3)
This response denies that Paula clearly uses ‘I’ as a device of self-reference. Rather,
she uses it to refer to the object in her chair that she counts as a person: her ut-
terances of ‘I’ s+-refer to P+, and s−-refer to P−. So it’s true on s+ that Paula uses
‘I’ to s−-refer to an object in the s−-extension of ‘person’. So it is true on s+ that
her utterance of ‘I am a person’ was true on s−. Generalising: it’s true on each
sharpening that her utterance of ‘I am a person’ was true on each sharpening.
Hawthorne captures the oddness of this view nicely:
“There is something exceedingly strange about a view according to which
. . . many people (perhaps most people) do not [clearly] have linguistic
devices of self-reference. Relatedly, it is extremely natural to think that
if a pronominal device has the conceptual role of the first-person pro-
noun in a person’s cognitive life, then that pronoun will be a device of
self-reference.” (Hawthorne, 2006a, p.190)
Maybe we can learn to live with this. But the analogous responses to the second
and third arguments are even worse: the person in my chair doesn’t clearly self-
refer using ‘I’. But that person is me. I certainly find this hard to believe, and I
26 Here we have another difference from Hawthorne’s puzzle: denying (2) blocks his argument
but not ours. The source of the difference is that, on the Sharpening View but not on epistemicism,
speakers aren’t one-one correlated with intended interpretations of their language. In particular,
they aren’t one-one correlated with intended referents for their own name.
Vague Reference 179
suspect that you do too for the corresponding problem for your use of ‘I’. We ought
not endorse so radical a view unless there really is no alternative.
3.3.2.4 Third response: deny (4)
This response denies that Paula’s utterances of the form ‘a is F’ are true iff ‘F’ ap-
plies to the referent of ‘a’, as Paula uses those expressions. Rather, Paula’s utter-
ances of that form are true iff ‘F’ applies to the referent of ‘a’, as we use ‘F’and ‘a’.27
Then it’s true on s+ that Paula’s utterance of ‘I am a person’ is true on s−; for as
we use ‘person’ (on s+), it applies to the object to which Paula/P+ refers using ‘I’,
namely P+.
This violates a foundational principle of good translation:
A correct translation of another’s utterances assigns them the truth-conditions
that their utterer used them to express.
By denying this principle, the present response amounts to an unacceptable form
of semantic chauvinism. It’s akin to denying the meaningfulness of French on the
grounds that English speakers do not use French vocabulary meaningfully. Why
should the near-coincidence of the person-candidates in Paula’s chair makes a dif-
ference to this? We should reject this response.
3.3.2.5 Fourth response: deny (5)
This response denies that both s+ and s− are intended interpretations of Paula’s
language. This will need finessing; for it is a datum that Paula’s language is vague.
The most promising strategy will vary the intended interpretations of Paula’s lan-
guage across intended interpretations of ours: s+ doesn’t count s− as an intended
interpretation. Then since it’s true on s+ that Paula’s utterance of ‘I am a person’ is
true on s+, it’s also true on s+ that Paula’s utterance was clearly true. Likewise for
s−. So it’s clearly true that Paula’s utterance was clearly true.
27 Care is needed here. For as we use ‘I’, it refers to ourselves, not to Paula. An exception is needed
for ‘I’: it’s Paula’s use that matters, not ours. Since (3) ensures that how Paula uses ‘I’ is determined by
how we use ‘Paula’, this gives the result we need. But this non-uniformity of truth-conditions should
already make us suspicious.
Vague Reference 180
This view makes it true on each interpretation that only one Paula-candidate
satisfies ‘person’ on any intended interpretation of her language. But Paula’s lan-
guage is our language. So it’s true on each interpretation of our language that only
one Paula-candidate satisfies ‘person’ on any intended interpretation of our lan-
guage. Likewise for all other people. And applying the analogous response to the
second and third variant puzzles: it’s true on each interpretation of my language
that only one Nick-candidate satisfies ‘person’ on any intended interpretation of my
language. It follows that ‘person’ and ‘Nick’ aren’t vague. Since Nick is the only
person in my chair, and Paula is the only person in her chair, and. . . , it follows that
names for persons are non-vague too, and hence that persons (and self-referrers
generally) cannot have vague boundaries.
If persons cannot have vague boundaries, why do they appear to? We must
either (i) explain away the mistaken appearance of vagueness, or (ii) adopt a non-
supervaluationist account of vagueness in the boundaries of persons. Both options
are problematic: why not apply this alternative account to all (apparent) vague-
ness? The supervaluationist should reject this approach to vague boundaries.
None of these responses is satisfactory. The Sharpening theorist therefore ought
not to treat vagueness in the boundaries of speakers and thinkers as vagueness
about which object speaks and thinks. In the interests of uniformity, they ought
not to treat any vague boundaries in this way.
3.3.3 Coincidence, time and modality
§1.4.1 presented several puzzles about the interaction of the Problem of the Many
with time and modality. They arose because Tibbles’s boundaries (i) become more
and less vague over time, and (ii) could have been more or less vague than they
actually are. This section argues that in order to solve these puzzles, advocates
of the Lewisian Proposal must endorse objectionable theses about the metaphysics
and semantics of temporal and modal discourse (§3.3.3.1). §3.3.3.2 explains why
these commitments are objectionable. Lewis himself held (versions of) these views:
the Lewisian Proposal about vague boundaries is inextricably linked with the rest
Vague Reference 181
of his metaphysical framework.
3.3.3.1 Coincidence, persistence and modality
This section argues from the Lewisian Proposal about vague boundaries to the dis-
junction of counterpart-theory with perdurance-theory. We’ve already rejected the
former (§1.1.2.1). We reject the latter in §3.3.3.2.
Consider the following world w in which Tibbles remains on the mat through-
out his life. Tibbles’s boundaries are precise until t1, when hair h becomes his only
borderline part: Tibbles-candidates T+ and T− nearly coincide after t1. Because
the following is clearly true, T+ and T− were on the mat before t1, when they were
coincident:
Tibbles was on the mat before t1.
Let v be a world just like w until t1, when Tibbles is annihilated. Tibbles’s bound-
aries are always precise in v. How many cat-candidates are on the mat in v? §1.4.1
argued that no answer is satisfactory.
Suppose there is only one cat-candidate on the mat in v. Four problems arise:
(a) Is the v-cat-candidate T+ or T−? Either answer seems arbitrary. This arbi-
trariness may be relieved by postulating a world u just like v except for which
Tibbles-candidate it contains. Since this is the only difference between v and
u, it follows that the events leading up to Tibbles’s birth were insufficient to
bring him into being.
(b) Suppose that v is actual. Then if Tibbles’s future boundaries had been more
vague than they actually are, then there would have been more cat-candidates
on the mat before t1 than there actually are. So the number of past objects
seems to depend on how the future unfolds.
(c) There is a dynamical sense of possibility in which what’s possible changes
over time: it used to be possible that I wouldn’t finish this thesis, but no
longer is. If there’s only one candidate in v, then it’s not possible in v for
Tibbles’s boundaries to become vague; for there’s only one interpretation on
which any possible future candidate was on the mat before t1 and hence no
Vague Reference 182
possible future in which there’s more than one intended referent for ‘Tib-
bles’. So it’s dynamically impossible in v for Tibbles’s boundaries to become
vague. More generally, it’s dynamically impossible in any world for Tibbles’s
boundaries to become more vague than they ever are.
(d) In w, it’s not clearly true that Tibbles would have existed had the boundaries
of the cat on the mat been less vague than they are. For in one of the closest
worlds to w in which the cat on the mat’s boundaries are precise, namely v,
one candidate doesn’t exist. So on any sharpening that assigns this candidate
to Tibbles, it isn’t true that Tibbles exists in all the closest worlds in which
the cat on the mat’s boundaries are less vague than they are in w.
Now suppose that both cat-candidates are on the mat in v, where they permanently
coincide despite belonging to the same kind. Two further problems:
(e) There’s no reason for both candidates to exist in v, other than that Tibbles’s
boundaries could have, but didn’t, later become vague. The number of objects
seems to depend on how things could have been.
(f) There need to be enough candidates on the mat in v to accommodate any
possible vagueness in Tibbles’s boundaries. So there will be many candidates
indeed, even in worlds where Tibbles’s boundaries are always precise.
Although I know of no responses to (e) and (f), there seem to be two responses to
(a)–(d) capable of granting that there’s only one candidate on the mat in v. The first
is counterpart-theory. The second is perdurance-theory.
Counterpart-theory decouples the modal and temporal profiles of ordinary ob-
jects from questions about their identity across times and worlds. The puzzles all
then concern the number of ways of selecting non-present/other-worldly counter-
parts, rather than the existence of candidates; questions about the number of can-
didates in a world can then be separated from questions about tracking them across
worlds. But since we rejected counterpart-theory in §1.1.2.2, we won’t examine this
response in detail. Instead, we proceed directly to perdurance.
Vague Reference 183
3.3.3.2 Perdurance
Perdurance is the version of temporal-parts theory that identifies ordinary persist-
ing objects, the referents of ordinary names, with sums of short-lived objects:28
If an object x exists at a time t, then there is something y such that (i) y exists
at and only at t, (ii) y overlaps at t exactly those things that overlap x at t, and
(iii) y is part of x. We say that y is x’s t-part.
On this view, cat-candidates are spatiotemporally extended “worms”. The v-candidate
is neither T+ nor T−, but the restriction of those two worms to times before t1,
when Tibbles’s boundaries became unclear.29 From this perspective, variation in
the extent of unclarity in Tibbles’s boundaries looks just like Lewis’s (1976) account
of fission and fusion: fusion occurs when worms that used not to share temporal-
parts come to do so; fission occurs when worms that used to share temporal-parts
cease to do so. The difference is simply a matter of scale. Let’s apply this to prob-
lems (a)–(d) before we reject it.
The perdurantist solution In response to (a): there’s no arbitrariness about which
candidate exists in v because neither does. Instead, the restriction of T+ and of T−
to times before t1, when they shared the same stages, exists in v. This restriction is
composed by their pre-t1-parts, each of which exists in v.
The response to (b) is a little more concessive. The number of past objects
does depend on how the future turns out, but only in a familiar and (supposedly)
unproblematic way. A spatial analogy is helpful.
The number of roads in a town depends on the world outside the town. What
appears within a town to be a single road might really be two roads that share a
(spatial) segment within the town and diverge outside of it. The number of roads
in a town depends on what happens elsewhere because roads can share some of
28 There are alternative formulations, but this captures the view’s core. For discussion see
Hawthorne (2006d) and Sider (2001a).29 We make the natural simplifying assumptions that (i) temporal parts have their temporal loca-
tion essentially, and (ii) spatiotemporal worms have their temporal parts essentially. It follows that
temporal extent is an essential property of spatiotemporal worms. Only our manner of presentation
turns on this.
Vague Reference 184
their spatial segments without sharing all of their spatial segments. The number
of roads within a region r thus depends on the relations between road-segments
entirely contained within r and road-segments elsewhere. But the number of road-
segments entirely contained within r is independent of their relations to road-
segments elsewhere. These features of roads are commonplace and unproblem-
atic. The perdurance-theorist appeals to analogous temporal claims in response to
problem (b).
The number of cat-candidates (worms) on the mat at a time depends on what
happens at other times. What appears at one time t to be a single candidate
could really be two candidates that share (temporal) t-segments, but not later t′-
segments. The number of candidates at t thus depends on the relations between
candidate-segments that exist only at t and candidate-segments that exist only at
other times. But the number of candidate-segments that exist only at t is inde-
pendent of their relations to candidate-segments elsewhen. The number of past
candidates therefore depends on the number of future candidates only in the same
unproblematic way in which the number of roads within a town depends on what
happens outside the town.30
As it stands, this solution to (a) and (b) is not fully general because it cannot
accommodate (c) and (d). Consider (c). Let T be the candidate-worm in v: T is
the only intended referent of ‘Tibbles’ in v. For any dynamical future possibility
u and object x in u, either x is identical to T or it isn’t. If it isn’t, then, in v, no
intended interpretation assigns x in u to ‘Tibbles’. So, in v, there’s at most one
intended referent for ‘Tibbles’ in any future dynamical possibility. So if Tibbles’s
boundaries ever aren’t vague, then they can’t become vague. Similarly for (d): at
most one w-candidate is identical to the v-candidate; so if each w-candidate is an
intended referent for ‘Tibbles’ in w, then it’s not clearly true in w that Tibbles is on
the mat in v.30 Does this really solve the problem? Perdurantism implies a formal analogy between the de-
pendence of (i) the number of past objects on the future, and (ii) the number of local objects on the
non-local. But it doesn’t follow that since (ii) is unproblematic, so is (i). It needs to be shown that the
analogy is more than formal, that (i) and (ii) are analogous in respect of being unproblematic. This is
just the point at issue.
Vague Reference 185
A modal analogue of perdurance is required:
If an object x exists at a time t in a world w, then there is something y such
that (i) y exists at, and only at, t in w, and (ii) y overlaps at t in w exactly those
things that overlap x at t in w, and (iii) y is part of x.31
Distinguish a modal-cat-stage wholly contained within v from a “modally extended”
cat-candidate. Cat-candidates are fusions of the temporally extended cat-worms
wholly contained within worlds. The existence of only a single such worm in v
therefore does not reduce the number of cat-candidates (partly) in v. These can-
didates provide many intended referents for ‘Tibbles’ both before t1 in v, and in
any dynamically possible future. This resolves (c). It also resolves (d) because the
candidate referents for ‘Tibbles’ in w are cross-world fusions which overlap objects
on the mat (that exist only) in v. Hence those candidates exist in v.
Modal perdurance seems committed to Lewis’s ontology of worlds as spatiotem-
porally and causally isolated concrete spacetimes. This is a very high cost; we
ought to reject modal perdurance unless we are Lewisian realists about possible
worlds. But without modal perdurance, and hence also Lewis’s modal ontology,
perdurance-theory is not a fully general solution to puzzles (a)–(d). However, there
are good reasons to reject perdurance even if we are prepared to accept Lewis’s
modal ontology.
Perdurance faces two major difficulties. The first is that it brings significant
complications in the semantics for object-language temporal discourse. The sec-
ond is that it is unclear whether we can understand the language in which the
perdurantist semantic theory is stated.
First problem for perdurance The key perdurantist thesis is that ordinary ob-
jects, the subjects of ordinary thought and talk, are spatiotemporally extended
worms. So suppose it was true yesterday that Tibbles was purring, but isn’t true
today. Under what conditions is an atomic predication like ‘Tibbles is purring’ true
at a time? The following won’t do:
31 Clause (i) implies that y only exists in w.
Vague Reference 186
‘a is F’ is true at t iff the referent of ‘a’ is F.
If the whole worm that ‘Tibbles’ refers to is F, then this truth-condition makes
‘Tibbles if F’ true yesterday iff it is also true today. This makes it impossible for
Tibbles to purr only temporarily. So the perdurantist needs a truth-condition like
the following:
‘a is F’ is true at t iff the t-part of the referent of ‘a’ is F.32
This makes it true yesterday that Tibbles was purring iff Tibbles’s yesterday-part
was purring, and not true today that Tibbles is purring iff Tibbles’s today-part is not
purring. The properties expressed by predicates like ‘is purring’ are thus primarily
properties of stages of persisting objects.
This truth-condition won’t do for all predications. Consider:
Tibbles is a cat.
This should be true whenever Tibbles exists. Applying the truth-condition above
gives:
‘Tibbles is a cat’ is true at t iff the t-part of the referent of ‘Tibbles’ is a cat.
But none of Tibbles’s t-parts is a cat: the ordinary persisting object Tibbles is a cat,
not any of his temporal stages. The perdurantist therefore faces a choice. The first
option is to deny that Tibbles himself is a cat. On this view, Tibbles is a cat in only
the following sense: each of his temporal-parts is a cat. The cost of this view is that
the subjects of ordinary thought and talk are not cats, clouds, chairs, humans etc.,
but fusions of momentary cats, clouds, chairs and humans.
The second option is to adopt a non-uniform account of temporal-modification.
On this view, some predications have the truth-condition above, and some have this
alternative truth-condition:
‘a is F’ is true at t iff the referent of ‘a’ is F.
It’s worth noting that this non-uniformity will arise anyway. Consider:
32 A slightly different truth-condition is the following: the referent of ‘Tibbles’ is such that its t-
part is F. The difference concerns whether the subjects or predicates of atomic predications vary over
time. The following discussion should be insensitive to this difference.
Vague Reference 187
Tibbles has been sleeping for four days.
Tibbles is a temporally extended worm.
Suppose these are both true today (as the second must be, in order for perdurance
to be true). Applying the first truth-condition gives:
The today-part of the referent of ‘Tibbles’ has been sleeping for four days.
The today-part of the referent of ‘Tibbles’ is a temporally extended worm.
Tibbles’s today-part has not been sleeping for four days. And Tibbles’s today-part is
not temporally extended. So the first truth-condition conflicts with our supposition
by making both sentences false. But applying the second truth-condition gives:
The referent of ‘Tibbles’ has been sleeping for four days.
The referent of ‘Tibbles’ is a temporally extended worm.
The temporally extended worm Tibbles may well have been sleeping for four days.33
And that worm is temporally extended. This second truth-condition therefore does
not make both sentences trivially false.
We can draw two morals from this. The first is that perdurance complicates our
semantic theory. The truth-conditions of some predications turn on the properties
of temporal-stages, while the truth-conditions of other predications turn on the
properties of worms themselves. The second moral is that this complication is
essential to the perdurantist theory; for without it, many of their key theses would
be false. If both types of predication were not present in natural language, then
the perdurantist would lack the expressive resources to state their view without
rendering it trivially false.
Second problem for perdurance It is not clear whether we can understand the
perdurantist semantic theory. They offer the r.h.s. of equivalences like the following
as an analysis of (the present truth-conditions of) the l.h.s.:
Tibbles is purring iff the now-part of Tibbles is purring.
33 Presumably, a worm has been sleeping for four days iff each of its temporal-parts over those days
is asleep.
Vague Reference 188
Tibbles is sitting iff the now-part of Tibbles is sitting.
For a large class of predications Fa, their perdurantist truth-conditions thus involve
applying F not to the referent of a, but to one of its temporal-parts. The difficulty
is that although we understand these predicates as they apply to ordinary objects,
it is not clear that we understand them as they apply to the temporal-parts of those
objects.
What does it mean to say that Tibbles’s today-part is purring? The natural an-
swer is: Tibbles is purring. The problem is that this explanation applies the pred-
icate ‘is purring’ to the persisting four-dimensional object Tibbles, rather than to
a stage of that object. It is therefore unclear whether the perdurantist can give an
account of the content of their proposed truth-condition for ‘Tibbles is purring’,
other than in terms of the properties of persisting objects. It is therefore unclear
whether they can provide an illuminating semantics for truths about persisting
objects in terms of truths about their temporal-parts. The suspicion is that when
a perdurantist applies an ordinary predicate F to a temporal-part x when stating
their semantic theory, the content of this statement is explicable only in terms of
the application of F to the persisting object of which x is a temporal-part. I do not
claim that this challenge cannot be met, but that it presents a significant obstacle
to an informative perdurantist semantic theory.
Perdurance-theory brings two difficulties. Firstly, it complicates our semantic
theory. Secondly, it is not obvious that we understand the predicates used in its
semantic theory (if the theory is supposed to be informative). We should there-
fore be reluctant to endorse the predurance-theoretic solution to problems (a)–(d)
above. Since we have also rejected counterpart-theory and no alternative solutions
are forthcoming, we should be reluctant to endorse the Lewisian Proposal about
vague boundaries that generates problems (a)–(d).
3.4 Conclusion
This chapter examined the Lewisian Proposal that vague boundaries result from
vagueness about which object’s boundaries are in question. We saw two ways of
Vague Reference 189
developing this idea in §3.1 and rejected one of them. Before examining the Pro-
posal itself, we defended the Sharpening View’s account of vague reference in §3.2.
With this account of vague reference in place, we examined three problems for the
Lewisian Proposal in §3.3. The first questioned whether it was a genuine solution
and exposed its widespread and radical hidden metaphysical assumptions. The
second suggested that the Proposal prevents an adequate characterisation of the
semantics of self-reference, and hence cannot be extended to cover objects capable
of such. The third committed the Lewisian to either counterpart-theory or perdu-
rance. This Lewisian Proposal was thereby revealed as an entrenched component
of Lewis’s metaphysical framework, not readily separable from the whole. Having
already rejected counterpart-theory in §1.1.2.1, we closed by rejecting perdurance.
In light of these problems, we should reject the Lewisian Proposal: Unger’s and
Lewis’s puzzles are neither sources nor symptoms of referential vagueness in our
names for ordinary objects. We must therefore consider vagueness in mereolog-
ical, constitutional and loctional concepts themselves, as those concepts apply to
ordinary objects. The next chapter defends a view of this kind.
190
Chapter 4
Identity Conditions and
Constitution
The previous chapter rejected Lewisian attempts to resolve the Problem of the
Many by postulating referential vagueness in our names for ordinary objects: Unger’s
puzzle does not induce unclarity about the referent of ‘Tibbles’, and vagueness
about Tibbles’s constitution is not a result of such unclarity. This leaves one option:
Tibbles’s boundaries are vague because mereological and locational vocabulary it-
self is vague, at least as it applies to ordinary objects. On this kind of view, Tibbles
is clearly the unique most cat-like object on his mat. Everything else thereabouts
is clearly not a cat, though it’s unclear of some things whether they constitute (or
compose or make up) a cat. None of Tibbles’s Many are cats, or even candidate
cats, but merely borderline cases of cat-constituters. Unger’s puzzle is one source
of this constitutional unclarity (though not uncontroversially of full-blown vague-
ness). This chapter develops a conception of ordinary objects that accommodates
this.
Mark Johnston and E.J. Lowe advocate similar views. §4.1 argues that their
views are unsatisfactory because they do not provide a unified solution to the puz-
zles, but merely an ad-hoc collection of theses designed to block the arguments for
many cats. §4.2 presents and develops the following thesis about ordinary objects:
persistence through change is explanatorily prior to constitution, mereology and
location. We argue from this thesis to the following solution to Unger’s puzzle:
Identity Conditions 191
Tibbles is constituted by each of the best candidates on his mat. §4.3 rebuts several
objections to this view. Our solution creates trouble for the idea that Tibbles “inher-
its” some properties, like mass, from his constituters: surely Tibbles doesn’t have
several (incompatible) masses. §4.4 develops an account of property-inheritance
that avoids this problem. We close with three accounts of vague constitution in
§4.5.
4.1 Vagueness in parthood and constitution
Mark Johnston (1992, §4) and E.J. Lowe (1995) defend views of the present kind.
Both distinguish constitution from identity and grant that there are many equally
good candidates on Tibbles’s mat. However, they claim, these candidates are clearly
not cats and clearly distinct from Tibbles; they are merely candidates to constitute
Tibbles the cat, not candidates to be him. Johnston and Lowe claim that it is vague
which candidate constitutes Tibbles, the one and only object on the mat with any
claim to be a cat.1
Lowe claims that this vagueness involves semantic indecision about the exten-
sion of ‘constitutes’, and should be handled supervaluationally. Johnston claims
that:
“The problem of the many simply shows that constitution is a vague re-
lation.. . . [O]n one legitimate sharpening [Tibbles] is constituted by one
of the [candidates], on another, another of the [candidates], and so on.
What is important for our purposes is that on no legitimate sharpening
is [Tibbles] identical with any one of the [candidates].” (Johnston, 1992,
p.100)
Johnston’s view is less than perspicuous. Calling constitution a vague relation
seems at odds with the linguistic conception of vagueness usually associated with
sharpenings and supervaluations.2 Note however, that Johnston and Lowe agree
1 Michael Tye (1996) also distinguishes Tibbles from his candidate constituters, though Tye is no
supervaluationist.2 Williamson (2003) discusses the relationship between supervaluationism and vague properties
and relations.
Identity Conditions 192
on the following: the candidates are clearly all non-cats, though it is vague which
constitutes Tibbles the cat. This section agues that this view is unsatisfactory as it
stands.
4.1.1 Why only one cat?
Johnston and Lowe do not provide a unified response to the Problem of the Many,
but an ad-hoc collection of theses united only by their role in blocking the argu-
ments for many cats. Their view is therefore unsatisfactory.
To see why Johnston and Lowe’s view is unsatisfactory, note that the iden-
tity/constitution and cat/constituter distinctions alone provide no reason to deny
that there are many cats on the mat. They are consistent with:
(1) Each candidate constitutes a cat.
This seems to follow from the following pair:
(2) The candidates are alike in respects relevant to their constituting cats (cat-
respects).
(3) Clearly, some candidate constitutes a cat.
Suppose x, y are alike in cat-respects and it’s clear that one of them constitutes a
cat. Then surely both must constitute cats; for otherwise they would differ in cat-
respects because only one would constitute a cat. So (1) seems to follow from (2)
and (3).
(1) alone doesn’t imply that there are many cats; for the candidates may all con-
stitute the same cat. But since no two candidates spatially coincide, the following
imply that no two of them constitute the same cat:
(4) If x constitutes some cat y, then x and y occupy exactly the same place.3
(5) No cat occupies more than one region at a time.
We now have a two-step argument from an abundance of candidates to an abun-
dance of cats. The first step concludes that each candidate constitutes a cat. The
3 We use ‘occupies’ to mean exact occupation: x occupies region r iff x fills and fits within r.
Identity Conditions 193
second step concludes that no two candidates constitute the same cat. The problem
for Lowe and Johnston is that they provide no reason to reject either step, or even
to endorse the vagueness of constitution. The identity/constitution distinction is
consistent with various responses to Unger’s and Lewis’s puzzles (including Lewis’s
own). The suspicion is that Lowe and Johnston’s solution does not constitute a uni-
fied theoretical package. To illustrate, consider the following two responses to our
two-step argument for many cats.
First response This response attacks the argument from (2) and (3) to (1): Tib-
bles’s candidates can be alike in cat-respects despite none clearly constituting a cat.
All that’s required in order for it to be clear that some candidate constitutes a cat,
Lowe and Johnston may claim, is that each candidate borderline constitutes a cat;
no candidate clearly does so, though none clearly fails to do so either. A superval-
uationist logic on which ∆∃xA doesn’t imply ∃x∆A is obviously congenial to this.
(2) and (3) imply that each candidate clearly constitutes a cat only if ‘constitutes’ is
not vague. The vagueness of constitution therefore invalidates the first step of our
argument for many cats.
There are three problems for this response. First, the identity/constitution dis-
tinction neither implies nor suggests that ‘constitutes’ is vague. It therefore pro-
vides no reason to believe that the argument is invalid. Second, even granting that
constitution is vague, Lowe and Johnston provide no reason to think that (1) is
false. Our argument shows that either constitution is vague or each candidate con-
stitutes a cat. No basis to prefer one disjunct to the other has been supplied. Third,
this disjunction is inclusive: it is consistent with the vagueness of constitution and
the identity/constitution distinction that many vaguely constituted cats are on the
mat. Lowe and Johnston provide no reason to think otherwise.4
Second response This response attacks the second step of our argument for many
cats. Since (4) and (5) seem too firmly embedded in our conception of the material
world to be plausibly denied, this response doesn’t reject, but modifies them. It
4 Matti Eklund (2008) raises a similar worry about appeal to ontological vagueness in response to
the Problem of the Many.
Identity Conditions 194
does so by denying that cats occupy space in the same way as their constituters:
cats occupy space by being constituted by objects that occupy space. The spatial
properties of cats are thus relational properties: cat x occupies region r relative to
its constituter y; this relation obtains iff y both constitutes x and occupies r in the
primary, non-relativised, sense.
On this view, (4) and (5) become:
(4′) If x constitutes some cat y, then: y occupies r relative to x iff x occupies r.
(5′) No cat occupies, relative to anything that constitutes it, more than one region
at a time.
(4′) is trivial, given the proposed account of spatial occupation by cats. (5′) follows
from the (plausible) assumption that no cat-constituter occupies more than one
region at a time. From the supposition that each candidate constitutes the same
cat—i.e. that each candidate constitutes Tibbles—(4′) implies:
For some distinct regions r, r′ and distinct candidates c, c′, Tibbles occupies r
relative to c, and r′ relative to c′.
Since this is consistent with (5′), it doesn’t follow that the candidates don’t all con-
stitute the same cat.
This second response suffers the same problem as the first. The identity/const-
itution distinction provides no reason to endorse it. That distinction doesn’t imply
that cats and their constituters occupy space in the fundamentally different ways
that this response claims they do.
In sum, then, although the identity/constitution distinction does create logical
space for (at least) two responses to the Problem of the Many, it provides no reason
to endorse either of them. It doesn’t even provide reason to deny that there are
many cats on Tibbles’s mat. Lowe and Johnston therefore do not provide unified
theoretical packages. This chapter seeks to do better. Our goal is a conception
of ordinary objects from which a solution to Unger’s and Lewis’s puzzles emerges
naturally, alongside the identity/constitution distinction, unclarity in constitution
and Tibbles’s uniqueness.
Identity Conditions 195
4.2 The Proposal
This section develops a conception of ordinary objects and applies it to Unger’s
puzzle of too many best candidates. §4.2.1 outlines the core idea. Central to this
idea are criteria of identity. §4.2.2 presents two kinds of identity criterion. §4.2.3
develops our proposal in accord with the first kind of criterion and applies it to
Unger’s puzzle. §4.2.4 does the same with the second kind of criterion. §4.2.5 then
argues that we needn’t choose between these proposals, and §4.2.6 closes by exam-
ining the concepts of matter and constitution employed throughout our discussion.
The next section turns to some objections.
4.2.1 Objects and change
Our primary interest is in ordinary objects, the kind of object singled out in or-
dinary thought and talk, and the inhabitants of the ordinary macroscopic world.
We’re also interested in the ordinary parts of these objects, e.g. the hearts and
lungs of animals, the cells of plants, and the legs of tables, because the Problem of
the Many arises for them too. So what is it to be an ordinary object?
Our proposal develops a broadly Aristotelian answer to this question: an ordi-
nary object is a subject of change; different kinds of object are subjects of different
kinds of change. This idea is captured by associating each ordinary kind K with
an identity condition that determines what changes Ks survive; identity conditions
determine the histories and futures of ordinary objects.
What is the content of our claim that to be an ordinary object is to be a subject
of change? A strong form of this thesis takes it as a (conceptual? metaphysical?)
analysis of belonging to the category of ordinary objects. However, the following
weaker thesis will suffice for our purposes whilst allowing us to avoid difficult
questions about just what such analyses amount to:
Identity conditions, history and change are explanatorily prior to the consti-
tution, mereology, and spatial location of ordinary objects.
The idea is that every fact about the constitution, mereology and location of or-
dinary objects is explicable in terms of the kinds of changes that those objects do
Identity Conditions 196
and don’t survive (in combination with other contingent facts). An object’s identity
condition determines its path through time and space; it’s constitution, mereology
and most other properties follow from what happens along this path.
David Wiggins defends a view along similar lines:
“Suppose I ask: Is a, the man sitting on the left at the back of the restau-
rant, the same person as b, the boy who won the drawing prize at the
school I was still a pupil at early in the year 1951? To answer this sort
of question is surprisingly straightforward in practice.. . . Roughly, what
organizes our actual method is the idea of a particular kind of contin-
uous path through space and time the man would have had to have
followed in order to end up here in the restaurant.. . . Once we have dis-
pelled any doubt whether there is a path in space and time along which
that schoolboy might have been traced and we have concluded that the
human being who was that schoolboy coincides with the person/human
being at the back of the restaurant, this identity is settled.. . . The conti-
nuity or coincidence in question here is that which is brought into con-
sideration by what it is to be a human being.. . . The contention is. . . that
to determine correctly the answer to our continuity question, the ques-
tion about the traceability of things through their life-histories, pre-
cisely is to settle it that, no matter what property φ is, a has φ if and
only if b has φ.” (Wiggins, 2001, pp.56–7)
I include the last sentence because one property of b is the property of being b; thus
to settle how to trace the histories of a and b is, in part, to settle whether a is b.
And how to trace the histories of a and b is settled by what kind of objects they are,
human beings in Wiggins’s example.
How does this help with the Problem of the Many? There’s more detail in
§§4.2.3–4.2.4, but a preview may be helpful. By determining what kinds of change
it survives, an object’s identity condition associates it with a path through space
and time. The Problem of the Many shows that a single object’s path may exhibit
a branching structure, passing through several nearly coincident spatial regions at
a single time, when examined on a sufficiently small scale. When this happens,
Identity Conditions 197
the ordinary object with this branching history is simultaneously constituted by
the occupants of each of these regions; these occupants are its Many. Our thesis
about the relative explanatory priorities of change and constitution even allows us
to argue for this last claim.
Let T be a cat on Tibbles’s mat at time t1. Let l2 be one of the candidates on
the mat at the later time t2. The question is: does l2 constitute T at t2? The answer
will be positive if the change c that T would have to survive in order to come to
be constituted by l2 at t2 is a kind of change that cats do survive. We may safely
assume that T does survive some change c∗, as a result of which it comes to be
constituted by some other candidate l∗2 at t2. Given how similar the candidates on
the mat at any given time are to one another—Unger’s puzzle only arises if l2 and l∗2
are alike in cat-respects—the changes c and c∗ will also be very similar: c and c∗ are
alike in respects relevant to whether they are the kind of change survived by cats.
But then, how could T survive c∗ but not survive c? If either is the kind of change
that cats survive, then surely they both are. Since, we assumed, T does survive c∗,
it also survives c. So T survives both changes, as a result of which it comes to be
constituted by both l2 and l∗2 at t2. Generalising: all the candidates on the mat at
t2 simultaneously constitute one and the same cat. A similar argument gives the
same result for t1. Hence T is, at every time, constituted by every candidate then
on the mat: T is Tibbles, the one and only cat ever on the mat.
The moral is that if objects are individuated by their histories rather than by
their microscopic constituents, then many equally good candidate cat-constituters
needn’t correspond to many cats. They may instead all constitute a single cat. The
Problem of the Many is a symptom of an overemphasis on mereology in the ontol-
ogy of ordinary objects.
To make good on these claims, more detail is required about identity criteria
and constitution. The next section introduces two kinds of identity criterion. Sub-
sequent sections use them to develop two versions of our proposal.
Identity Conditions 198
4.2.2 Two kinds of identity criterion
Three putative examples of identity criteria are prominent in the literature; those
for sets, directions and cardinal numbers:5
(∀x : Set(x))(∀y : Set(y))(x = y↔ ∀z(z ∈ x ↔ z ∈ y))(Extensionality)
∀x∀y(d(x) = d(y)↔ x‖y)(Dir)
∀F∀G(#(F) = #(G)↔ F1–1G)(HP)
‘Set’ is a predicate true of exactly the sets; ‘d’ denotes the function from lines onto
their directions, and ‘‖’ the relation of parallelism; ‘#’ denotes the function from
concepts or properties onto their cardinalities, and ‘1–1’ the relation of one-one cor-
respondence. Thus (Extensionality) says that sets are identical iff they have exactly
the same members; (Dir) says that lines have the same direction iff those lines are
parallel; and (HP) says that concepts have the same cardinality iff those concepts
are in one-one correspondence.
Abstracting away from these specific examples (and ignoring the higher-order
quantifiers in (HP)), there are two kinds of statement here:
One-level (∀x : F(x))(∀y : F(y))(x = y↔ R(x, y)).
Two-level ∀x∀y( f (x) = f (y)↔ R(x, y)).
The labels are from Williamson (1990, §9.1). In a true one-level criterion, R is an
identity condition for Fs. In a true two-level criterion, R is an identity condition
for f (x)s.6 The formal properties of identity require that identity conditions are
equivalence relations.
One-level and two-level criteria obviously have different logical forms. But for
our purposes, the question of logical form is secondary to another difference. In
a one-level criterion, R is a relation on the very objects whose identity is at issue
on the left. In a two-level criterion, by contrast, R need not be. Witness that coex-
tensiveness is a relation on sets, though parallelism and one-one correspondence
5 (∀x : A) is a quantifier restricted to satisfiers of A.6 We use ‘identity criterion’ for statements of either of the forms above, and ‘identity condition’
for the relations on the r.h.s. of true identity criteria. We’ll also call (putative) identity conditions for
Fs one-level or two-level, according as to whether they are relations on Fs or not.
Identity Conditions 199
are relations on lines and concepts respectively, not on directions and cardinalities.
Hence the choice between one- and two-level criteria amounts to the following: is
the identity condition for Ks a relation on Ks?
Actually, this isn’t quite right. An identity condition R for Ks in a two-level
criterion may be a relation on Ks. The key difference between the two kinds of
statement is that the logical form of a two-level criterion for Ks does not require
that R be a relation on Ks, whereas the logical form of a one-level criterion does.
For simplicity, we’ll confine ourselves to genuinely two-level criteria in the sequel:
two-level criteria in which the putative identity condition R is not a relation on
f (x)s.
We want to use identity criteria to explicate the idea that an ordinary object
is a subject of change: the identity condition R for Ks is explanatorily prior to the
constitution, mereology and locations of Ks. The choice between one-level and two-
level criteria thus amounts to a choice concerning which kinds of change to regard
as prior to constitution, which kinds of change explain the the facts about consti-
tution. A one-level version of our proposal concerns changes in ordinary objects
themselves. A two-level version concerns changes in something else, whose prop-
erties and relations are systematically correlated with identity amongst ordinary
objects.
We’ll ultimately prefer a one-level view, though we won’t reject two-level crite-
ria outright. Note however that either kind of criterion will suffice. Our proposed
solution to the Problem of the Many is insensitive to this nuance of formulation.
The explanatory priority of change and history over constitution and mereology is
what’s doing the work. The following sections exhibit this by developing both one-
level and two-level versions of our proposal and applying them to Unger’s puzzle.
4.2.3 The one-level proposal
The one-level variant of our proposal claims that, for each ordinary kind K, some
unique equivalence relation RK on Ks satisfies:
(L1-K) (∀x : Kx)(∀y : Ky)(x = y↔ RK(x, y))
Identity Conditions 200
Applied to cats, we get:
(L1-Cat) (∀x : Cat(x))(∀y : Cat(y))(x = y↔ Rc(x, y))
Cat x is identical to cat y iff x bears Rc to y. Rc thus determines what changes cats
survive. What kinds of change are these? Two issues arise. Firstly, what are the
subjects of these changes? Secondly, what kinds of change are these? We address
these in turn.
Since (L1-Cat) is a one-level criterion, Rc is a relation on cats. It holds between
a cat x and a cat y iff the changes that x would have to survive in order to be y are
of the kind that cats do survive. The one-level proposal thus prioritises changes in
cats themselves—as opposed to changes in their underlying matter or the regions
they occupy—over the constitution and mereology of cats. Rc bears on the survival
of cats through changes in their matter only insofar as those material changes cor-
relate with changes in cat(s). This addresses the first question at the end of the
previous paragraph. So let’s turn to the second: what kinds of change do cats sur-
vive?
Set to one side the question of whether a complete and informative account of
Rc is possible. We’ll return to that in §4.3.5. Our grasp on Rc is firm enough for
present purposes, regardless of whether English contains a (possibly complex) ex-
pression coextensive with it. Cats persist through hair loss, purring, falling asleep,
pouncing, digesting and countless other ordinary and familiar kinds of change.
They don’t survive drowning, starvation, squashing and the like. Note that these
are all primarily macroscopic changes; changes whose subjects are cats.
Unger’s puzzle arises because our concept ‘cat on the mat’ fails to determine
a unique microscopically individuated portion of matter as the constituter of the
ordinary macroscopic object in question. Likewise, our concept ‘the loss of hair h’
fails to determine a unique microscopically individuated occurrence as the mate-
rial basis for the ordinary macroscopic change in question. Our one-level proposal
puts this to work in a solution to Unger’s puzzle.
Identity Conditions 201
4.2.3.1 Applying the one-level proposal
This section applies the one-level proposal to Unger’s puzzle of too many can-
didates. For simplicity, we’ll assume that cats are constituted by lumps of mat-
ter and restrict attention to those lumps on Tibbles’s mat that are candidate cat-
constituters. These notions of matter and constitution are elaborated in §4.2.6.
Let T be a candidate lump of matter on the mat at t1. Suppose for simplicity that
it’s the only candidate then on the mat, and hence that it then constitutes Tibbles.
Let T2, T∗2 be two candidates lumps on the mat at the later t2. Suppose for simplicity
that they’re the only candidates then on the mat. Suppose Tibbles clearly survives
from t1 to t2: the changes that occur on his mat over that duration aren’t the kind
that destroy cats. We want to show that Tibbles is constituted by both T2 and T∗2 at
t2.
Since Tibbles survives from t1 to t2, we know that he survive some change c, as
a result of which he comes to be constituted by at least one of T2, T∗2 at t2. Suppose
that it’s T2. Now consider the change c∗ that Tibbles would have to survive in order
to come to be constituted by T∗2 at t2. How could Tibbles survive c but not c∗? Since
T2 and T∗2 are very similar, c and c∗ are too. If it’s implausible that only one of T2
and T∗2 constitutes a cat, then surely it’s also implausible that only one of c and c∗
is the kind of change that cats survive. The ordinary changes and process that cats
survive though—the loss of a hair, purring, walking, pouncing and so on—don’t
seem to distinguish between objects as similar as T2 and T∗2 in respect of which
comes to constitute the cat that participates in them as a result of its undergoing
that change. Both c and c∗ are equally good candidates to be changes that Tibbles
survives. Since he clearly survives one of them, surely he must survive both. Hence
T2 and T∗2 both constitute Tibbles at t2.
Nothing in this argument turned on the assumption that T is the only candidate-
constituter on the mat at t1. So we can generalise: for any candidate x on the mat
at any time t, and for any candidate y on the mat at any other time t′, x constitutes
the same cat at t as y does at t′.
This argument shows that any pair of candidates drawn from different times
constitute the same cat as one another. It doesn’t follow that there’s only one cat on
Identity Conditions 202
the mat; for all we’ve said so far, each of the candidates may constitute more than
one cat. T, for example, may constitute two cats at t1, one of which is constituted
by T2 and t2, and one of which is constituted by T∗2 at t2. Similarly, even if T2 and T∗2
both constitute the same cat, they may both constitute more than one. We’ll now
argue that this isn’t the case.
Our key thesis is that change is explanatorily prior to constitution. One aspect
of this is that all the facts about constitution are explicable in terms of facts about
change. So if any candidate constitutes more than one cat, then that’s explicable in
terms of the kinds of change that cats survive. If there’s no such explanation, then
no candidate constitutes many cats; in which case there’s only one cat on Tibbles’s
mat. What would such an explanation be?
An explanation in terms of change for there being more than one cat on the
mat would involve a pair of changes such that, from the occurrence of both on the
mat, it follows that there’s more than one cat; a pair of changes that couldn’t both
be survived by a single cat. Ordinary changes like the loss of a hair don’t seem to
provide this. The only candidates I can find are pairs like:
The change a cat undergoes when it ceases to be constituted by T and comes
to be constituted by T2 and not by T∗2 .
The change a cat undergoes when it ceases to be constituted by T and comes
to be constituted by T∗2 and not by T2.
The supposition that a cat survives both these changes is inconsistent. So no cat
can do so. So if cats do survive these changes (and both occur on the mat), then
there were two cats on the mat at t1, both constituted by T. This kind of suggestion
faces four problems.7
Firstly, it’s not obvious that these are genuinely one-level changes because they’re
only specifiable using features of matter. Secondly, since both changes are speci-
fiable only using constitutional vocabulary, they conflict with our thesis of the ex-
7 This objection doesn’t require the claim that cats do survive these changes, or that they do occur
on the mat. All that’s required is that we can’t rule out that they do, and hence can’t rule out there
being many cats.
Identity Conditions 203
planatory priority of change over constitution.8 Thirdly, these changes are so sim-
ilar that it’s unclear how any cat could survive only one of them. But since no
cat can survive both changes, it follows that no cat survives either; neither is the
kind of change that cats survive. Fourthly, these changes appear too artificial and
gerrymandered to be plausibly taken as characteristic of cats.
Given these difficulties, there seems no explanation in terms of change for how
any candidate could constitute more than one cat. So no candidate does. Tibbles
is therefore the only cat ever on the mat. At each time, he’s constituted by each
candidate then on the mat.
This one-level proposal employs the phenomenon that generates Unger’s puz-
zle as part of a solution. That puzzle arises because our ordinary sortal concepts
don’t make sufficiently fine-grained distinction amongst lumps of matter in respect
of which lumps constitute the satisfiers of those concepts. Likewise, our sortal con-
cepts don’t distinguish between the changes that cats would have to survive in
order to come to be constituted by some one later candidate in preference to any
other. Since those changes have equally good claim to be survived by Tibbles and
he survives at least one of them, Unger’s reasoning concludes that he survives them
all. The result is that Tibbles comes to be multiply constituted by all the best can-
didates on his mat.
4.2.4 The two-level proposal
Recall the general form of a two-level criterion:
∀x∀y( f (x) = f (y)↔ R(x, y))
Two-level criteria treat the f (x)s as invariants across R-connected series of xs.
Since our interest is in ordinary objects, the f (x)s will be ordinary objects. Thus or-
dinary objects are invariants across R-connected series of some other kind of entity.
What kind of entity?
8 One might respond to these first two problems by claiming that although the changes in question
are only specifiable in terms of constitution and features of matter, this is simply a function of the
expressive limitations of natural language, rather than a reflection of any deep metaphysical fact.
But the onus is on our opponent to justify this perspective. Why should we believe that appearances
are misleading in the way that they claim?
Identity Conditions 204
The quantifiers in our two-level criteria range over a kind of entity such that, no
matter how unlike one another members of that kind may be, they “support” the
same ordinary object iff they stand in R. Changes in such entities that don’t bear
on R, don’t bear on their relations to ordinary objects. Two questions arise. Firstly,
what are these entities? And secondly, what is this “support” relation?
Our concern is with the constitution of cats by the lumps on Tibbles’s mat. It’s
therefore natural to take the domain of f as lumps of matter, its range as cats,
and the relation between x and f (x) as the constitution relation. On this view,
an ordinary object is an invariant across changes in various lumps of matter, its
constituters over time. Our proposal then becomes the thesis that variation in the
properties of matter is explanatorily prior to the constitution and mereology of
ordinary objects; the r.h.s. of a two-level criterion is explanatorily prior to the left.
This won’t quite do because ordinary objects are typically constituted by differ-
ent matter at different times. So we’ll take the quantifiers in our two-level criteria
to range over ordered pairs 〈l, t〉 of lumps l and times t, and take f as the (par-
tial) function from 〈l, t〉 pairs onto the object (if any) constituted by l at t. Then
we can simulate cross-time variation in the relata of R without having to compli-
cate the criteria by introducing extra argument places for times on the right. We
will however frequently write as if R held amongst different lumps at (and across)
times, and as if fc mapped different lumps onto different cats at different times.
The difference is merely notational.
Two problems remain. Firstly, different kinds of object survive different kinds
of change in their matter. Secondly a single lump may constitute several objects
at once (e.g. some wool may constitute a thread and a jumper). The following
proposal avoids both problems.
Let K be any ordinary kind. Let fK be the (partial) function from 〈l, t〉 pairs
onto the K (if any) that l constitutes at t. Then some unique relation RK satisfies:
(L2-K) ∀x∀y( fK(x) = fK(y)↔ RK(x, y))
Applied to cats:
(L2-Cat) ∀x∀y( fc(x) = fc(y)↔ Rc(x, y))
Identity Conditions 205
Rc determines a range of paths through various lumps of matter over time. Any
lumps that lie on these paths constitute the same cat (whilst they lie on that path).
Rc thereby determines the survival of cats through changes in their matter. Before
applying this to Unger’s puzzle, following are six comments by way of elucidation.
First comment Although (L2-K) makes it formally possible to identify Ks with
equivalence classes of RK-related 〈l, t〉 pairs, it does not mandate doing so. Neither
should we wish to. Cats are ordinary concrete objects, not set-theoretic constructs.
They do not have members, they are not abstract, and (according to our proposal)
they are not “formed” from or dependent upon lumps of matter, times or 〈l, t〉
pairs. An equivalence class of RK-related 〈l, t〉 pairs is, at best, a formal model of a
cat.
Second comment The identity/constitution distinction follows directly from fo-
cusing on genuinely two-level criteria. It also follows from the fact that fK maps
distinct 〈l, t〉 pairs onto the same K: since cats are constituted by different lumps at
different times, constitution is not identity; cats are not identical to their matter.
Third comment Although fK was introduced as the function from 〈l, t〉 pairs onto
the K constituted by l at t, this is not a definition of fK. We want to leave it open
how closely fK coincides with our ordinary notion of constitution. (§4.2.6 discusses
this notion.) In particular, we want to leave it open whether a single ordinary
object might be simultaneously constituted by several lumps of matter. In part,
that’s because our solution to Unger’s puzzle requires that they can be. But it also
coheres with our thesis of the explanatory priority of change and persistence over
constitution.
Fourth comment The fundamental difference between (L1-Cat) and (L2-Cat) con-
cerns whether Rc, the identity condition for cats, is a relation on cats or on lumps
of matter. (L2-Cat), unlike (L1-Cat), treats cats as invariant through changes in
non-cats, specifically in their matter. On this view, the changes undergone by cats
bear on their persistence only insofar as they correlate with changes in their matter.
Identity Conditions 206
One benefit is that the two-level proposal avoids the appearance of bootstrapping
that may attend the one-level solution.
Fifth comment Peter Simons (2000, 2008) also endorses two-level identity crite-
ria for ordinary objects (and continuants more generally). His view differs from
ours in three respects. (i) Simons takes the identity condition for Ks as a relation
on events, not on matter or other continuants. (ii) Simons takes the notion of an
invariant with more metaphysical seriousness than we do. His two-level criteria
reflect a view of continuants as abstractions from, and dependent upon, metaphys-
ically fundamental events. Our two-level proposal involves no theses about the
non-fundamentality of ordinary objects, or their dependence on the domain of fc.
We claim only that the relations on that domain are what determine the survival of
cats. (iii) Simons endorses a different account of property-possession to our own,
but we won’t go into that here.
Sixth comment E.J. Lowe writes:
“[T]he parallelism of lines can provide a criterion of identity for the di-
rections of lines only because directions are ontologically (and indeed
conceptually) dependent on lines in a way that lines are not on direc-
tions. But this immediately raises a difficulty for anyone seeking to ex-
tend [two-level criteria] to names of what we might, in an Aristotelian
vein, call (primary) substances, since these (assuming they exist) are pre-
cisely the objects standing in no such relationship of ontological depen-
dency to other objects.” (Lowe, 1989, p.4; original italics)
If there are any substances, cats are presumably amongst them. And our concep-
tion of ordinary objects is closely akin to the Aristotelian notion of substance. So
Lowe’s objection had better be unsound. It is.
Grant for the sake of argument (the highly dubious assumption) that there is a
contentful notion of a specifically ontological kind of dependence. Our two-level
proposal involves no explicit theses about such dependence. It merely posits (i) a
structural thesis connecting cat-constitution and a relation on matter, and (ii) the
Identity Conditions 207
priority of identity conditions over constitution and mereology. Lowe’s objection
therefore carries weight only given an argument from our proposal to the relevant
claims about dependence. Lowe offers no explicit argument, but the following is
suggestive:
“[A]n acceptable criterion of identity for φs. . . should reveal, in an in-
formative way, what φ-identity ‘consists in’ (to use Locke’s well-worn
phrase). In particular, then, it should reveal to us wherein φ-identity
differs from ψ-identity.” (Lowe, 1991, p.193)
Suppose that if K-identity consists in a relation on Fs, then Ks are ontologically
dependent on Fs. It follows that if cats are ontologically independent, then cat-
identity doesn’t consist in a relation on non-cats. So if Lowe’s claim about identity
criteria is correct, then (L2-Cat) is not an acceptable identity criterion for cats.
If sound, the argument in the previous paragraph shows that (L2-Cat) does not
provide an analysis of cat-identity, or an account of what cat-identity consists in.
This doesn’t undermine the two-level view for two reasons. Firstly, we could regard
the r.h.s. of (L2-Cat) as an analysis of its l.h.s. without regarding Rc as an analysis
of cat-identity; for in the l.h.s. ‘ fc(x) = fc(y)’, the identity sign is flanked by com-
plex terms formed using functional signs that denote a constitution relation, and
variables ranging over matter. If the two-level theorist presents Rc as an analysis of
anything, it’s as an analysis of this complex statement, not bare identity. Secondly,
the two-level theorist is not compelled to regard the r.h.s. of (L2-Cat) as an analysis
of anything. That criterion may instead be taken to contribute to a better under-
standing of the kind cat, identity, constitution and material change by exposing
their inter-connections: (L2-Cat) imposes a structure on these connections, which
information about Rc imbues with content. Lowe employs too strong a notion of
ontological dependence if the independence of ordinary substances is incompatible
with systematic correlations between their persistence and the properties of mat-
ter. Surely the existence of such connections is uncontroversial. Lowe’s objection
therefore fails.
Now the two-level proposal is in place, let’s apply it to Unger’s puzzle.
Identity Conditions 208
4.2.4.1 Applying the two-level proposal
We want to use (L2-Cat) to show that there’s never more than one cat on Tibbles’s
mat. This section presents two strategies. The first requires slightly more substan-
tive metaphysical assumptions than the second.
First strategy We begin by arguing that if there’s ever n cats on the mat, then
there’s never more than n.
Suppose there are n cats on Tibbles’s mat at t1: 〈l1, t1〉, . . . 〈ln, t1〉 each bears Rc
to something, though not to one another. Suppose also that there are n + 1 cats
on the mat at some other time t2: 〈l∗1 , t2〉, . . . 〈l∗n+1, t2〉 each bears Rc to something,
though not to one another. Since there are more l∗s than ls, some pair 〈l∗i , t2〉 either
(i) bears Rc to the same 〈lj, t1〉 pair as some other 〈l∗k , t2〉 pair, or (ii) bears Rc to no
〈lj, t1〉 pair. But (i) implies that l∗i and l∗k constitute the same cat at t2; in which case
there aren’t n + 1 cats on the mat at t2, contrary to our second supposition. And (ii)
implies that l∗i constitutes a cat that wasn’t on the mat at t1, which, we may assume,
is obviously false: the occurrences on Tibbles’s mat didn’t bring a cat into being or
fuse Tibbles with some cat not previously on his mat. So our two suppositions are
incompatible. So if there are ever n cats on the mat, then there are never more than
n cats on the mat.
So far, so good. But we want to show that there’s never more than one cat on
Tibbles’s mat. This follows from:
Independence Whether Rc holds between 〈l, t〉 and 〈l∗, t∗〉 is independent of what
happens outside of the duration from t to t∗. Whether l constitutes the same
cat at t as l∗ does at t∗ turns only on what happens between t and t∗.
Possible Uniqueness Either (i) events could possibly unfold so that there’s only
one candidate on Tibbles’s mat at some point in the future, or (ii) there’s a
possibility w in which (a) there’s only one candidate on Tibbles’s mat and (b)
events could possibly unfold from w to give the actual present situation.
To understand Possible Uniqueness, think of reality as a branching structure. Dif-
ferent branches represent different possible complete histories (including futures).
Identity Conditions 209
Shared portions of branches represent shared portions of possible histories. Pos-
sible Uniqueness says that each branch intersects with one where there’s only one
candidate on Tibbles’s mat: each history shares a portion with one where there’s
only one cat on Tibbles’s mat; whatever the situation on Tibbles’s mat, it could
either lead to a future situation or be the result of a (merely possible) past situa-
tion in which there’s only candidate on the mat.9 Unger’s argument suggests that
such one-candidate situations are unlikely to be actual, not they are impossible.
The argument above shows that any branch with only one candidate at some point
has only one cat at all points. Independence then extends this to any connect-
ing branch. Possible Uniqueness says that every branch connects with some such
branch. So there’s never more than one cat on Tibbles’s mat.
Although Possible Uniqueness and Independence seem natural, they are non-
trivial and I know of no direct arguments for them. In support of Possible Unique-
ness, it’s hard to imagine instances of Unger’s puzzle that couldn’t be reached from,
or couldn’t unfold into, some situation in which there’s only one candidate: what
would Tibbles have to now be like in order to prevent there ever being a time at
which there’s only one candidate? And in support of Independence, we might claim
that Rc marks an intrinsic similarity between its relata. For those suspicious of
these theses however, a slightly different strategy is available.
Second strategy The Problem of the Many arises only when the candidates are
alike in respects relevant to cat-constitution. In a two-level setting, this amounts
to the candidates being alike in respects relevant to Rc. Since Rc is an equivalence
relation, each candidate bears Rc to itself. So if candidate x doesn’t bear Rc to
candidate y, then they differ in at least one respect relevant to cat-constitution:
only y bears Rc to y. Since they don’t differ in such respects, each candidate bears
Rc to each other. Similarly, if x but not y bears Rc to some future-candidate z, then
x and y differ in a respect relevant to cat-constitution, namely, the bearing of Rc
9 A weaker claim that would suffice is: either (i) each history shares a portion with one in which
there’s only one cat; or (ii) each history shares a portion with a history that shares a portion with
one in which there’s only one cat; or (iii) each history shares a portion with a history that shares a
portion with a history that shares a portion with one in which there’s only one cat; or (iv) each history
shares. . . .
Identity Conditions 210
to z. Since the candidates are alike in those respects, any two present-candidates
bear Rc to the same future-candidates. Since Rc is an equivalence relation, they
also bear it to one another. So they constitute the same cat, and there’s never more
than one cat on the mat. At every time, the cat on the mat is constituted by each
best candidate-constituter on the mat.
Two features combine here. The first is that the candidates are alike in cat-
respects. The second is that we’ve fleshed out cat-respects using a relation. This
second feature forces us to consider a range of candidates—those to which a given
candidate bears Rc—when assessing resemblance in cat-respects. The first then en-
sures that the candidates bear Rc to the same ranges of candidates. The similarities
between candidates that give rise to Unger’s puzzle thereby contribute to resolving
it.
A slightly different way to see this last point is as follows. Suppose that there
are two earlier candidates T1, T2 and two later candidates T3, T4. If Rc holds from,
say, T1 to T3 but not to T4, then this provides an example of a change in T1’s matter
that cats don’t survive. But the similarity of T3 to T4 makes this implausible if that
similarity is close enough to undermine the claim that only one of them constitutes
a cat. So T1 and T4 constitute the same cat iff T1 and T3 do. Since T1 constitutes the
same cat as either T3 or T4, it must constitute the same cat as them both. Likewise
with T2 in place of T1. So T1 and T2 stand in Rc, and therefore constitute the same
cat. Generalising: there’s never more than one cat on the mat, and it’s constituted
at each time by each candidate then on the mat.
4.2.5 How many levels?
We’ve now got two responses to Unger’s puzzle in place. Isn’t this one too many?
Which should we choose? Although §4.3.5 presents some (inconclusive) reasons to
prefer the one-level proposal, the two views aren’t competitors.
Our core thesis is the explanatory priority of change over constitution. The
one-level and two-level views differ over which kinds of change they claim take
priority. But the one-level view needn’t claim that changes in cats take priority
over changes in their matter. And the two-level view needn’t claim that changes
Identity Conditions 211
in matter take priority over changes in cats. We might instead see both kinds of
change as explanatorily prior to constitution. This requires that one-level and two-
level identity criteria don’t deliver conflicting results about the persistence and
constitution of cats. But since there’s no reason to think that they will, we’re not
forced to decide between these two views.
4.2.6 Matter and constitution
We’ve made free use of the idea that ordinary objects are constituted by lumps of
matter. Can we be sure that lumps of matter exist? And what is this constitution
relation? This section addresses these questions in turn.
We needn’t assume the existence of lumps of matter in any significant sense.
Maybe fundamental reality consists of pluralities or aggregates of microscopic par-
ticles, regions of spacetime, or something else entirely. Our use of ‘matter’ is best
understood as a placeholder for whatever physical stuff ordinary objects are ulti-
mately made out of. And if there’s no absolutely fundamental stuff, then an appro-
priate choice of level will suffice. Or we could simply re-parse our discussion in
terms of the occupation of spatiotemporal regions by objects.
Our use of ‘constitution’ also comes with minimal theoretical baggage. It’s best
seen as a technical term for an ordinary concept. It’s beyond doubt that ordinary
objects are in some sense made from other (smaller) entities. We use ‘constitution’
to denote whatever relation occupies this “making up” role, without adopting any
substantive metaphysical views about it, other than that several entities can bear
this relation to a single ordinary object (though see §4.5.1). Our key thesis is that,
whatever relation constitution is and whatever ordinary objects are constituted by,
persistence and change are explanatorily prior to constitution and mereology.
4.2.7 Identity criteria: concluding remarks
This section presented two solutions to Unger’s puzzle of many best candidates.
Although their details differ, the core idea is the same. Each ordinary kind K is
associated with an identity condition that determines what changes its members
survive. Unger’s puzzle arises when these changes don’t determine a unique lump
Identity Conditions 212
of matter as the best candidate to constitute a given K. Since these lumps are all on
a par and at least one constitutes a K, they all do. But because they’re so similar—
they’ve all been selected as K-constituters by the identity condition of an individual
K—these lumps all simultaneously constitute one and the same K.
We can now diagnose the fundamental flaw in the argument from many lumps
to many cats: it assumes that objects are built up from or individuated by their
microscopic constituents in an ontologically significant sense. On this kind of view,
to be an ordinary object (of a kind K) is to be made out of smaller objects in an
appropriate way. This makes it inevitable that differences in suitably arranged
small objects correlate with differences about which ordinary objects they make
up. Unger’s puzzle arises when many such suitably arranged collections almost,
but not quite, coincide.
Jettisoning this conception of objects invalidates the argument from many can-
didates to many cats. We’ve also seen that it brings positive arguments against
there being many cats. An overemphasis on mereology in contemporary ontology
disguises this fundamental mistake.
Classical extensional mereology, and variants thereof, provide prominent ex-
amples of this flawed conception of the relationship between ordinary objects and
their microscopic parts.10 But the mistake is not confined to these views. Fine
(1999, 2008) proposes a hylomorphic view according to which an object’s funda-
mental nature—its real definition, or essence—is given by a list of its parts and the
form, or universal, they instantiate To be a cat, for example, is to be a collection
of atoms in the form of a cat. Although Fine’s approach is radically unlike classi-
cal mereology, it shares the same flaw. If an ordinary object’s underlying nature is
given by a list of its microscopic constituents, then different lists must correspond
to different objects. The inclusion of a form as an additional item on the list makes
no difference to this. The lesson of Unger’s puzzle is that any conception of objects,
including Fine’s hylomporhism, that associates each ordinary object with a unique
collection of microscopic constituents is false. To be an ordinary object is not to be
made out of appropriately arranged stuff, but to survive through certain sorts of
change.
10 See Simons (1987) for discussion of classical mereology.
Identity Conditions 213
4.3 Objections
This section responds to seven objections to our proposal. §4.3.1 considers two
objections arising from the principle of Unique Constitution. §4.3.2 addresses an
objection we raised against a similar proposal in §1.1.4.2. We discuss a purported
similarity between the Problem of the Many and fission and fusion in §4.3.3, and
Lewis’s scepticism about the cat/cat-constituter distinction in §4.3.4. §4.3.5 asks
whether an informative statement of either kind of identity criterion is possible.
§4.3.6 closes by examining the claim that identity criteria attempt the impossible,
namely an analysis of identity.
4.3.1 Unique Constitution
Consider:
Unique Constitution (UC) No cat can ever be constituted by more than one lump
of matter at a time.
Our solution is incompatible with UC. This section considers two problems this
creates.
4.3.1.1 First objection
The first objection runs as follows: since UC is true and incompatible with our pro-
posal, that proposal is false. We should not find this compelling. Simply asserting
a theoretical claim like UC without supporting argument carries no suasive force.
Following are two arguments for UC (and responses).
The first argument for UC appeals to ordinary usage of constitutional vocab-
ulary. The idea is that UC is a deeply entrenched part of ordinary discourse, and
therefore carries strong intuitive support. Evidence comes from our use of definite
descriptions, e.g.: ‘the clay that used to make up the statue is now a set of dishes’.
Unless the statue was constituted by a unique lump of clay, the initial description
in this example is improper and the whole sentence therefore untrue. But, we may
assume, surely it was true; such claims are part of what fix the meaning of ‘con-
stitutes’. Our response to this argument must await §4.5.1. That section shows
Identity Conditions 214
that our proposal doesn’t make definite descriptions like ‘the matter of Tibbles’
improper and can therefore accommodate this kind of linguistic consideration.
The second argument for UC comes from §4.1. Cats are located wherever their
constituters are located. Cats are also only located in one place at a time. But
only one lump can occupy a place at a time.11 So cats cannot be simultaneously
constituted by multiple lumps.12
There are two responses to this argument. One must await §4.4. There we
provide an account of Tibbles’s location and other inherited properties that is com-
patible with our proposal. The other response can be stated now: our proposal
was not simply that change is prior to constitution, but that it is also prior to loca-
tion. Exact parallels of our arguments for multiple constitution show that Tibbles
multiply occupies the regions occupied by each of his best candidate-constituters.
Although our experience of reality shows that cats don’t simultaneously occupy
many quite disparate regions, that experience is silent about regions that differ as
little as those occupied by Tibbles’s Many.
4.3.1.2 Second objection
This second objection claims that our solution tacitly assumes that UC is false; it
only follows from our proposal that Tibbles is constituted by all of the candidates
because we’ve built it in from the start. If this is right, then our proposal is no more
theoretically unified than those of Johnston and Lowe that we criticised in §4.1.
This objection fails. Although the truth of UC does block our arguments for
multiple constitution, we needn’t assume that it’s false. We can instead remain ag-
nostic about UC, and see whether consideration of the changes survived by cats tell
for or against it. Our arguments for multiple-constitution show that these consid-
erations tell against UC. The agnostic who seeks justification for UC in the changes
that cats survive will not find one. Given our thesis that facts about constitution
are explicable in terms of facts about change, it follows that UC is not true.
11 Recall our use of ‘occupy’ for exact occupation: x occupies r iff x fills and fits within r.12 Even if several lumps can occupy a single place, the first two premisses alone conflict with our
solution; for we claim that Tibbles is constituted by lumps that occupy different places at the same
time.
Identity Conditions 215
4.3.2 Cats and maximal lumps
§1.1.4.2 considered the suggestion that Tibbles is constituted by the largest cat-like
lump on his mat: the cat-like lump that includes all the cat-like lumps that include
it. We rejected this because there’s no reason to believe that there’s a unique such
a lump. We then considered a variant suggestion: Tibbles is the fusion of all the
largest cat-like lumps on his mat. We rejected this for two reasons. Firstly, there’s
no reason to believe that this fusion will itself be a cat-like lump.13 Secondly, this
identifies the property of constituting a cat with the property of being a fusion of
near-coincident largest cat-like lumps. But since Unger’s puzzle already concerned
the property of constituting a cat, this simply changes the subject.
How does this differ from our proposal? Aren’t we claiming that Tibbles is con-
stituted by the fusion of every sufficiently cat-like lump on his mat? If so, then these
objections tell against our proposal too. Luckily, this isn’t what we’re claiming.
We claimed that Tibbles is constituted by each candidate on his mat; he is multi-
ply constituted by all the candidates, not uniquely constituted by their fusion. Our
claim is that the changes cats survive aren’t fine-grained enough to distinguish be-
tween many candidates, not that they’re sufficiently fine-grained as to distinguish
the fusion of those candidates from everything else (other than indirectly, as the
fusion of the candidates these changes don’t distinguish amongst). The underlying
logical point is that R(x1, y), . . . , R(xn, y) don’t imply R(x1 + . . . + xn, y), where ‘+’
denotes a fusion operation.
4.3.3 Fission and fusion
The extent of Unger’s puzzle varies over time. At the level of matter, this variation
is very similar to fission and fusion. But one thing we really don’t want to say
about fission is that the post-fission lumps both constitute one and the same object.
Consider, for example, microbial reproduction, brain-transplant cases, or cutting a
plant in half and re-planting the results. Two microbes, two people and two plants
13 The following jointly imply that the fusion of Tibbles’s nearby largest cat-like lumps isn’t a cat-
like lump: (i) they are largest cat-like lumps; (ii) there are many such lumps. Suppose that this fusion
is cat-like. Since it includes all the other cat-like lumps, it is the unique largest cat-like lump. But
this contradicts (ii).
Identity Conditions 216
are clearly the result, not one bi-located microbe, person or plant. Shouldn’t we say
the same about Unger’s puzzle?
There is a significant difference of scale between the two puzzles. Fission in-
volves much greater change than does an increase in the extent of Unger’s puzzle.
Although the two cases may begin in a similar manner, fission-products typically
occupy non-overlapping regions and have independent futures. The changes in-
volved in fission are too great for the object in question to survive (or so great that
they only indeterminately survive or. . . ). Variation in the extent of Unger’s puzzle
over time is a much smaller change than fission, one that the objects in question
do seem to survive; the candidates that result don’t have independent futures or
occupy entirely disjoint regions of space. Despite sharing a superficially similar
structure, the difference of scale between fission and Unger’s puzzle justifies our
treating them differently.
4.3.4 Lewis on cats and cat-constituters
Lewis rejects the distinction between cats and their constituent lumps:
“[E]ven granted that Tibbles has many constituters, I still question
whether Tibbles is the only cat present. The constituters are cat-like in
size, shape, weight, inner structure, and motion. They vibrate and set
the air in motion – in short, they purr (especially when you pat them).
Any way a cat can be at a moment, cat-constituters also can be; anything
a cat can do at a moment, cat-constituters also can do. They are all too
cat-like not to be cats. Indeed, they may have unfeline pasts and fu-
tures, but that doesn’t show that they are never cats; it only shows that
they do not remain cats for very long.” (Lewis, 1993a, p.168)
This is unpersuasive.
Firstly, possession of an unfeline past or future does show that something isn’t a
cat. Tibbles never was and will never be a scattered object, though his constituent
lump probably was and will be again. So Tibbles is distinct from his constituent
Identity Conditions 217
lump.14
Secondly, there are many ways a cat can be that its constituent lump(s) cannot.
Cats, for example, can have the identity conditions and modal profile of a cat; but
no lump can. Neither can lumps of matter purr; although matter can vibrate and
set the air in motion, it’s far from obvious that they, rather than the cats that they
constitute, thereby come to purr. Fine (2003) provides similar examples: a statue
may be Romanesque or well-made, but the bronze from which it is fashioned can-
not. These categorial differences seem to be a reasonably well-entrenched feature
of ordinary usage. Lewis could simply reject these differences; ordinary usage may
be misleading. But this is a cost, and certainly not a theoretically neutral response
to the arguments for the cat/constituter distinction.
So, there are good reasons to distinguish ordinary objects from their constituent
matter. These reasons aren’t unassailable, but neither are they without force. With-
out strong positive reason to reject these differences, which Lewis has not provided,
his scepticism about the cat/cat-constituter distinction carries little weight.
4.3.5 Stating the criteria
Can we give an informative statement of either a one-level or two-level criterion of
identity for cats? I certainly don’t know how to. Does this tell against our proposal?
This section argues that it doesn’t.
4.3.5.1 Stating a two-level criterion
Applied to cats, the two-level proposal claims that a unique relation Rc satisfies:
(L2-Cat) ∀x∀y( fc(x) = fc(y)↔ Rc(x, y))
where fc is the (partial) function from 〈l, t〉 pairs onto the cat (if any) that l con-
stitutes at t. The question now arises: what is Rc? The only answer I can supply
14 Lewis (1993a, pp.167–8) complains that the distinction between cats and their constituters is
unparsimonious and unnecessary, given that material change can be accommodated by temporal-
parts theory. But we rejected perdurance in §3.3.3.2 and the counterpart-theoretic semantics needed
for stage-theory in §1.1.2.1: temporal-parts theory is an incorrect account of constitutional change.
Identity Conditions 218
is:15
Rc is the relation that holds between 〈l, t〉 and 〈l∗, t∗〉 just in case l constitutes
the same cat at t as l∗ does at t∗.
There are two problems with this account of Rc. Both follow from our account of
temporally relativised constitution: x constitutes y at t iff fc(x, t) = y. For then our
statement of (L2-Cat) amounts to:
∀x∀y( fc(x) = fc(y)↔ fc(x) = fc(y))
The first problem is that this is not a two-level criterion: Rc is a relation on the
fc(x)s, not on the xs. The second problem is that it is obviously uninformative.
How severe are these problems? That depends on whether the two-level theo-
rist is committed to providing an informative statement of (L2-Cat). They are not.
Note first that our argument against many cats didn’t require an account of Rc.
Our solution to Unger’s puzzle relies not on the content of Rc, but on the structural
relationship between Rc, constitution and cat-identity captured by (L2-Cat).
There is no reason to expect that any single English word will be coextensive
with Rc. There may (though, equally there may not) be an English disjunction, each
of whose disjuncts corresponds to one (type of) instance of Rc. But the two-level
view provides no reason to think that such a disjunction would be finite. Hence it
provides no reason to think that we can give informative expression to Rc.
Our inability to give a non-trivial statement of (L2-Cat) reflects the relative po-
sitions of ordinary objects and matter within our cognitive architecture. We can
know about and refer to matter only because we can know about and refer to the
objects it constitutes: a portion of matter is accessible to us primarily as the matter
of a cat, or a dog, or the top half of a trout and bottom half of a turkey, and so on.
We can know that cats survive certain change in their matter only because those
15 The difficulty in stating a two-level criterion is reinforced by the observation that the identity
condition for Ks determines their existence condition also: x exists iff something is identical to x. An
account in terms of the properties and relations of matter of the conditions under which a cat comes
into being and all the possible changes that it can and can’t survive would be a very impressive
accomplishment.
Identity Conditions 219
changes correlate with changes in cats themselves (which we know them to sur-
vive). This doesn’t refute the two-level view because (a) that view doesn’t concern
our cognitive, epistemic or linguistic access to cats, but the relationship between
change and constitution, and (b) we can regard the trivialising account of Rc as us-
ing a relation on cats to fix the referent of an expression for a relation on matter; the
semantic value of ‘Rc’ is fixed in just the same way as that of any other theoretical
term (Lewis, 1970b).
The two-level theorist (who is not also a one-level theorist) does however re-
gard the relative cognitive, epistemic and linguistic priorities of cats and matter as
misleading. They do not reflect the underlying priority of material change over the
constitution and persistence of cats. This reversal of priorities is a cost. Is there any
positive argument for regarding these relative priorities as misleading, and hence
also for accepting this cost? Not that I know of.
Another difficulty is that the two-level view threatens to undermine our ability
to know about the persistence of cats. Our judgements about the identity and diver-
sity of cats are based on the properties and relations of cats, not on the properties
and relations of their matter. On the (pure) two-level view however, these relations
bear on cat-identity only insofar as they systematically correlate with relations on
the matter of their relata. Were there no such correlations, we couldn’t know about
the persistence of cats. A more attractive approach would more closely connect the
basis on which we make judgements about cat-identity with that which determines
the correctness of those judgements. This is what the one-level view offers.
We’ve highlighted three commitments of the two-level version of our proposal:
(i) The cognitive and linguistic priority of the persistence of cats over the proper-
ties and relations of their matter is misleading; relations on matter determine
the survival of cats.
(ii) The identity condition Rc for cats is a theoretical posit, expressible only using
terms for cats and constiutional vocabulary like ‘the matter of. . . ;
(iii) Our judgements about the persistence of cats are informed by a relation that
bears on their persistence only indirectly, via correlations with relations on
matter.
Identity Conditions 220
The one-level view faces none of these commitments.
4.3.5.2 Stating a one-level criterion
The one-level view fares better than the two-level view, as regards an informative
account of the identity conditions for cats. Applied to cats, that proposal claims
that a unique relation Rc satisfies:
(L1-Cat) (∀x : Cx)(∀y : Cy)(x = y↔ Rc(x, y))
Can we express Rc in a manner that renders (L1-Cat) informative? As with the
two-level criterion, I certainly don’t know how to. We can however state some
informative sufficient conditions for the survival of cats, thereby illuminating Rc:
cats survive through the loss of their hairs, beginning and ceasing to purr, pouncing
on and eating mice, and myriad other familiar kinds of change. We perceptually
track cats along certain paths through space and time by tracking these kinds of
change. If one of these paths connects cat x with cat y, then x = y. (Recall the
quote from Wiggins on p.196.) Our ability to isolate these paths and to track cats
along them indicates a grasp on Rc—we track cats by tracking the changes that they
undergo, not changes in their matter or anything else—even though our language
lacks the resources to express it in a manner that makes (L1-Cat) informative.16
The idea is that we don’t perceive a bare identity relation on cats, but know about
identity amongst (and hence the persistence of) cats because we grasp a relation
that at least approximates to Rc and which is, in typical cases, equivalent to identity
amongst cats.
Given some scene-setting, we can also state some informative necessary condi-
tions on the survival of cats, e.g.: cat x is identical to cat y only if cat x walked,
16 Isn’t a sufficient condition for the identity of cat x with cat y given in terms of paths through
space and time more appropriate for a two-level criterion, than a one-level criterion? It depends on
how the locations on the path are specified, and what unites those locations into a cat-survival path.
The one-level theorist will conceive it as the path combining the location of cat x1 at time t1, with the
location of cat x2 time t2, with. . . . These locations are united because the cat x1 survived a change as
a result of which it was in the location of x2 at t2, and x2 survived a change as a result of which it was
in the location of x3 at t3, and so on. The locations are thereby united by the survival of cats through
change.
Identity Conditions 221
crawled, pounced or leaped across the room, to occupy the location that cat y
now occupies. Here we need enough scene-setting to rule out human intervention,
earthquakes throwing x across the room, and so on.
The one-level theorist’s account of Rc is thus radically unlike the two-level theo-
rist’s. There seems little prospect for an accurate description in terms of the proper-
ties of matter of even the most simple and mundane changes that cats survive, like
walking across a room. Accurate one-level descriptions however seem utterly un-
problematic. The one-level theorist’s difficulty with giving informative expression
to Rc lies in combining these individually informative descriptions of its instances
into a single statement, rather than, as with the two-level theorist, expressing them
using appropriate vocabulary.
On the one-level view, we can describe, or even point to, instances of Rc. This
allows us to give various informative partial accounts of that relation. We may not
be able to convert this into an informative and exhaustive account of Rc, but since
our account of Rc is not exhausted by its role in (L1-Cat), that principle needn’t be
seen as entirely trivial.
Neither the one-level nor two-level view is refuted by our inability to give an
informative and exhaustive statement of the identity conditions for ordinary ob-
jects. We have however, seen that the two-level view brings costs that the one-level
view does not. If we must choose between them, then, other things being equal,
we should choose the one-level view. But since we needn’t make a choice, the most
satisfying view is probably one that combines both proposals.
4.3.6 Identity and analysis
This section considers two objections to the idea that an identity criterion is an
analysis of identity. Williamson (1990, pp.144–5) presents versions of both (though
he’s clear that they apply only to one-level views). Hirsch (1982, ch.3 §1) presents a
version of the second. We’ll argue that neither objection undermines our proposal
because our one-level and two-level identity criteria aren’t intended as analyses of
identity in other terms.
Identity Conditions 222
The first objection is that no analysis of identity is possible. The concept of
identity is so fundamental and basic to our conceptual scheme that any attempt to
analyse it in other terms is guaranteed to fail. A related complaint is that identity
is utterly simple, and therefore unanalysable.
The second objection is that the notion of identity is univocal. The same identity
relation holds amongst Ks as holds amongst K∗s. Different kinds don’t have differ-
ent identity criteria because if they did, then identity amongst Ks and amongst K∗s
wouldn’t be univocal.
The two-level view is immune to both complaints. The identity sign on the
left of a two-level criterion is flanked by complex terms formed using functional
signs and variables ranging over matter. The content of the l.h.s. therefore goes
beyond bare identity. This additional complexity means that (i) the l.h.s. may be
analysable, even if identity isn’t, and (ii) the l.h.s. may differ in content between
the criteria for Ks and K∗s, even if the notion of identity doesn’t. So let’s consider
one-level criteria.
The first objection assumes that an identity criterion for Ks should provide an
analysis of the identity relation, as it holds amongst Ks. Our one-level theorist
may reject this assumption. The identity condition for Ks is supposed only to de-
termine what changes Ks survive, not to determine what identity amongst Ks is.
Furthermore, since we’ve granted that only a partial account of Rc may be possi-
ble, we can hardy be accused of offering an analysis of identity amongst cats. But
another worry may arise: what theoretical interest do identity conditions hold, if
only partial and incomplete accounts of them are possible? The answer is that
they contribute to a better understanding of identity, the kind K, constitution and
change by exposing some of the ways in which those concepts interact: the pro-
vision of necessary and sufficient conditions is not the only route to philosophical
understanding.17
This also undermines the second complaint. Since our identity criteria aren’t
intended as analyses of identity in other terms, cross-kind variation in identity
17 Wright (1999, §6) and Wiggins (2001, Preamble, §10) describe similar approaches to philosoph-
ical analyses of truth and identity respectively. Horsten (2010) develops a similar role for identity
conditions in particular.
Identity Conditions 223
conditions is compatible with the univocity of identity. It follows only that different
kinds of object persist through different kinds of change.
A residual worry may remain: if identity is absolutely simple, then even a par-
tial elucidation in other terms will be impossible. The worry is baseless. One-level
identity conditions aren’t equivalent to identity, but to the restriction of identity to
the ordinary kind K. Even if no elucidation of identity is possible, a partial account
of the identity conditions for K may nonetheless contribute to our understanding
of the kind K and its role in delimiting a domain of objects.
In short, a more holistic conception of philosophical analysis combines with the
limited scope of our proposal to undermine both objections to one-level identity
criteria.
4.4 Property-possession
We’ve argued from the priority of change over constitution to the thesis that Tibbles
is constituted by each of the best candidates on his mat. Before turning to Lewis’s
puzzle of vague constitution, we should say a little more about this proposal. In
particular, we need to address the question: what properties does Tibbles have?
There is, after all, more to be said about Tibbles than what he is constituted by and
when; he has a mass, colour, location and sometimes purrs.
4.4.1 Three kinds of property-inheritance
Ordinary objects inherit properties from their constituters. For example, Tibbles
has a particular mass, location and colour because he’s constituted by something
with that mass, location and colour.
Ordinary objects don’t inherit all their properties: cats purr, but lumps of mat-
ter don’t; people act and think, but lumps of matter don’t; statues are beautiful,
Romanesque or well-made, but pieces of clay aren’t. This isn’t a rejection of sys-
tematic connections between these non-inherited properties of objects and those of
their constitutors, just of their direct inheritance. The following discussion should
be understood as restricted to inherited properties.
Identity Conditions 224
Were Tibbles constituted by exactly one lump, his property-inheritance would
be relatively unproblematic.
Naïve Inheritance (NI) Tibbles has φ iff Tibbles’s constituter has φ.
Multiple constitution makes the description ‘Tibbles’s constituter’ improper. NI
therefore fails to settle anything about Tibbles’s properties. An alternative is needed.
Two natural candidates are:
Universal Inheritance (UI) Tibbles has φ iff each of his constituters has φ.
Existential Inheritance (EI) Tibbles has φ iff at least one of his constituters has φ.
A third option modifies the logical form of the connection between Tibbles and his
properties, by relativising it to his constituters:
Relativised Inheritance (RI) Tibbles has φ relative to x iff x both constitutes Tib-
bles and has φ.
We assess these in turn. UI and EI will be rejected in favour of RI. Three versions
of RI will then be developed, and two defended.
4.4.2 Against Universal Inheritance
Suppose that Tibbles’s constituters don’t all have exactly the same mass. UI implies
that Tibbles doesn’t have any particular mass, despite being constituted by lumps
that do. Likewise for spatial location. But then in what sense is he a material object?
One might respond that although Tibbles doesn’t have any particular mass or
location, he is massive and he is located because each of his constituters is massive
and located. But surely it’s analytic that something is massive or located only if it
has some particular mass or particular location. It is obscure what being massive or
located might amount to, if not the possession of some particular mass or location.
So we should reject this response.
As well as undermining Tibbles’s status as a material object, UI undermines
our ability to know about him. How can we causally interact with something that
doesn’t have a spatial location or mass? And if we can’t causally interact with Tib-
bles, then it’s unclear how we can know about him. Although we can causally in-
teract with his cat-like constituters, none of those constituters is Tibbles. So causal
Identity Conditions 225
interaction with them doesn’t alleviate the problem of how we can know about
Tibbles.
In light of these two problems, we should reject UI.
4.4.3 Against Existential Inheritance
Suppose that Tibbles’s constituters don’t all have exactly the same mass. EI implies
that Tibbles has many masses. But since distinct masses are incompatible determi-
nates of the same determinable, this is impossible. At best, EI requires substantial
modifications to our ordinary conception of property-incompatibility and the de-
terminate/determinable contrast. We therefore reject it.
4.4.4 In defence of Relativised Inheritance
Only RI remains. On this view, Tibbles has a range of particular masses and loca-
tions, each relative to one his constituters. So unlike UI, RI doesn’t deprive Tibbles
of having a particular mass and colour. And since Tibbles doesn’t have incom-
patible masses simpliciter, but only relative to different constituters, RI, unlike EI,
doesn’t conflict with our ordinary conception of property-incompatibility.
An argument of sorts from our core thesis to RI is possible. According to our
proposal, Unger’s puzzle shows that, when examined closely enough, the changes
Tibbles survives supply him with a branching path through space and time. No
single branch contains the whole of his history, though each is one of his histories.
So when Tibbles’s inherited properties differ across branches, no single assignment
of those properties to him can tell the whole story. In order to say everything there
is to say about Tibbles’s properties, we have to say what branch they are found on.
RI implements this idea.
This section elaborates RI by defending it against an objection. Two defensible
forms of the view will be found, though one is preferable to the other.
4.4.4.1 A problem for RI
§§4.4.4.2–4.4.4.4 examine three responses to the following objection to RI: accord-
ing to RI, Tibbles has few, if any, intrinsic properties; in particular, he has no mass,
Identity Conditions 226
location, shape, or any other intrinsic property that he should inherit from his con-
stituters.
Without supporting argument, we should be unmoved by this objection. Fol-
lowing is one such argument.
According to RI, the sense in which Tibbles has a mass differs from the sense in
which a lump has a mass; for Tibbles has a mass relative to a lump, while lumps
simply have mass. Tibbles’s having a mass therefore involves his bearing a relation
to things with that mass, while a lump’s having a mass does not. We can now adapt
an argument of Lewis’s:
“I protest that there is. . . nothing in the picture that has [5kg] simpliciter.
. . . Instead of having [5kg] simpliciter, [Tibbles] bears the [having-relative-
to] relation to it and [a lump l]. But it is one thing to have a property,
it is something else to bear some relation to it. If a relation stands be-
tween you and your properties, you are alienated from them.” (Lewis,
2002, p.5. Lewis is objecting to theories that explain intrinsic change
by treating instantiation as a relation between objects, properties and
times. We’ve modified his example to fit the present case.)
There are two claims here:
(6) 5kg is an intrinsic property of Tibbles.
(7) Having an intrinsic property is not a matter of bearing a relation to it.
We’ve already got:
(8) If Tibbles has φ only relative to a constituter, then his having φ is a matter of
his bearing a relation to φ (and that constituter).
Together these imply that Tibbles doesn’t have 5kg only relative a constituter. Gen-
eralising: ordinary objects don’t possess their intrinsic properties only relative to
their constituters. So RI is false.
How might we respond? Premisses like (6) concern the paradigms that fix the
content of our notion of intrinsicality. The goal is a theory of properties and intrin-
sicality that is compatible with such claims. So the defender of RI must reject (7)
or (8). Two kinds of resistance to (7) are possible:
Identity Conditions 227
Having any property is a matter of bearing a relation to it.
Although having some properties is not a matter of bearing a relation to them,
the having of inherited properties by ordinary objects is.
These are examined in §4.4.4.2 and §4.4.4.3 respectively. Only the latter is defen-
sible. §4.4.4.4 presents an attack on premiss (8):
Having a property relative to a constituter need not be a matter of bearing a
relation to that property (and constituter).
Although the second attack on (7) is defensible, we’ll see that this third option is
preferable.
4.4.4.2 An instantiation relation?
This section considers rejecting (7) on the grounds that all property-possession is
analysable into the bearing of a relation between object and property. Bradley’s
Regress shows that this view is not tenable: some structures cannot be analysed
into the bearing of a relation amongst their constituents.18
To see this, consider the thesis:
Relational Analysis of Instantiation (RAI) Whenever an object x1 bears a rela-
tion R to objects x2, . . . , xn, this is analysable into the bearing of an instantia-
tion relation I amongst R, x1, . . . , xn.19
Suppose that a1 bears R to a2, . . . , an. By RAI: this is analysable into the bearing of
an instantiation relation I amongst R, a1, . . . , an. By RAI: this last fact is analysable
into the bearing of an instantiation relation I′ amongst I, R, a1, . . . , an. By RAI:
this is itself analysable into the bearing of an instantiation relation I′′ amongst
I′ I, R, a1, . . . , an. And so on ad infinitum. The output of each analysis by RAI is a
suitable input for analysis via RAI: each relational fact is the first element of an
18 Our mention of structures doesn’t bring commitment to entities that are structures. There may
be no structures, but only entities that are (collectively) structured. Talk about structures is for
convenience only.19 Treat monadic properties as the limiting case of relations.
Identity Conditions 228
infinite series of further relational facts; the obtaining of any fact in this series is
analysed into the obtaining of its successor.
Two problems arise, one metaphysical and one epistemological. The metaphys-
ical problem assumes that, in a good analysis, the analysans is more metaphysically
fundamental than the analysandum. RAI then implies that (a) every relational fact
decomposes into an infinite series of ever more fundamental relational facts. If the
defender of RAI finds series’ of this kind problematic, then they might respond by
excluding some elements of the series from the domain of RAI. This implies that
(b) there is a seemingly arbitrary point in the series at which the analysis termi-
nates. Let us grant for the sake of argument that the notion of x being more meta-
physically fundamental than y—conversely: y ontologically depending on x—is
contentful. Even given this dubious assumption, neither (a) nor (b) is obviously
problematic. Hence Bradley’s Regress is not obviously a metaphysical problem.
Consider (a). Although it sounds problematic, it is unclear what argument might
be brought against infinite sequences of ever more fundamental facts. This isn’t to
say that such sequences aren’t problematic, only that it’s unclear what could justify
a view either way.
Consider (b). What is the relevant notion of arbitrariness? Well, the terminus of
our series of ever more fundamental relational facts would presumably be utterly
fundamental: its obtaining cannot be analysed into the obtaining of any further
relational fact. So there is no account in other terms of why such a series terminates
where it does. So the existence of a terminus in our series of relational facts is not
itself objectionable. And neither is the belief that there is such a terminus, given an
argument against infinite chains of ontological dependence. It would be arbitrary
to believe, of any element in the series, that it is the terminus. But since that isn’t
what (b) requires, it’s no reason to find (b) objectionable. It’s unclear what further
argument for a problematic form of arbitrariness might be appealed to.
Since neither (a) nor (b) is obviously problematic, RAI’s relational analysis of
instantiation is not obviously a metaphysical problem. It is better seen as an episte-
mological problem. The following draws on Fraser MacBride’s (2005a) discussion,
though it’s unclear whether he would agree with our conclusion.
The (a?) point of analysis is explanation. One ought to endorse an analysis
Identity Conditions 229
therefore only if one has good reason to believe that the analysans can explain the
analysandum. One ought to endorse RAI therefore, only if one has good reason to
believe that the bearing of a relation I amongst R, x1, . . . , xn can explain R’s holding
amongst x1, . . . , xn. But since RAI applies also to I’s holding amongst R, x1, . . . , xn,
one must also have good reason to believe that the holding of a relation I′ amongst
I, R, x1, . . . , xn can explain this fact. And since RAI applies to I′s holding amongst
I, R, x1, . . . , xn, one must also have good reason to believe that. . . . At each stage, the
phenomenon being explained is the very phenomenon used in the explanation: the
bearing of a relation. One therefore ought not to endorse RAI unless one already
had good reason to believe RAI successful; RAI should not come to be accepted by
those who don’t already believe it. In the absence of such prior belief, we should
reject RAI. Bradley’s Regress shows that one can never acquire reason to believe
RAI unless one already has reason to believe it.
Since we shouldn’t believe RAI, we shouldn’t believe that entering into the rela-
tional structure R(a, b) by its constituents R, a, and b is analysable into the bearing
of a relation amongst those constituents. Some structure is not analysable into the
bearing of relations.
The situation resembles an argument with the global sceptic. The sceptic’s po-
sition is consistent and very difficult, if not impossible, to refute. The question is
whether we should join them in that position. Unless we are already sceptics, there
seems no reason to do so. And unless we are already committed to RAI, there seems
no reason to believe it, or even to believe that the kind of explanation it posits could
be successful.
If this is correct, then there is a difference between possessing a property and
bearing a relation to it. If Tibbles “has” 5kg either by being related to it, or by being
related to something else that has it, then he does not really have 5kg (or, at least,
does so only in an extended sense defined in terms of the primary sense). The next
section considers a different objection to (7) that grants this conclusion, but denies
that this difference is problematic.
Identity Conditions 230
4.4.4.3 A relational account of inheritance
Let us grant the conclusion of the previous section: if Tibbles has a mass-property
only relative to a constituter, then he has that property only in an extended sense,
not the primary sense in which his constituter has it. Is this an objectionable differ-
ence in sense? If not, then we may reject thesis (7) from p.226, and the argument
against Relativised Inheritance along with it.
Here is one flawed reason to think that this difference is not objectionable.
When Tibbles has 5kg in the extended sense of being related to something that
has 5kg in the primary sense, the relation in question is constitution: Tibbles is con-
stituted, or made out of, something that really does have 5kg. The flaw is that it’s
unclear why this should help.
The concern about RI was that if it is true, then Tibbles does not really have
any mass-property. Since Tibbles is distinct from each of his constituters, what
mass-properties they have is irrelevant to this concern. This is not alleviated by
calling the relation between Tibbles and a lump ‘constitution’. One reason is that
we are using ‘constitution’ only as a place-holder for whichever relation occupies
the pre-theoretic making-up role (§4.2.6). There is a more powerful reason also. If
constitution brought elimination or reduction, then this would alleviate the prob-
lem. But it does not: cats are neither reducible to, nor eliminable in favour of,
their matter. What’s needed is some kind of metaphysically substantial notion of
constitution, on which constituted and constituter, although distinct, are less than
completely distinct; a halfway house between elimination and non-elimination.20
It is far from clear that there is any such notion. And even if there is, the position
that results is in tension with our claim that change is prior to constitution: an ob-
ject’s underlying nature is not given by listing its microscopic constituents and the
way in which they are put together, but by specification of the changes it survives.
So let us set this view aside.
A more promising strategy asks why we should deny that Tibbles’s having 5kg is
a matter of his being related to a constituter that has 5kg in the primary sense. Why
20 Claims like the following only add to the obscurity: constituted entities are no increase in being
beyond, or an ontological free lunch given, the existence of their constituters.
Identity Conditions 231
should that be objectionable? Lewis’s work on change provides the only answer that
I know of.
How is intrinsic change possible? Nothing can be both bent and straight; yet
many things are bent at one time and straight at another. What role do the times
play here? One answer is that intrinsic properties like bent and straight are really
relations to times; they are really relations, bent-at- and straight-at-, between objects
and times. Against this view, Lewis claims:
“As we persist, we change. And not just in extrinsic ways, as when a
child was born elsewhere and I became an uncle. We also change in
our own intrinsic character, in the way we ourselves are, apart from our
relationship to anything else.. . . When I change my shape, that isn’t a
matter of my changing relationship to other things, or my relationship
to other changing things. I do the changing, all by myself.” (Lewis,
1988a, p.187)
But if Lewis’s having a shape is a matter of his being constituted by something with
that shape, then his changing shape is a matter of his changing relations to other
changing things. So, Lewis will conclude, his having an intrinsic property like a
shape or a mass is not a matter of his being constituted by something with that
shape or mass.
MacBride (2001, §2) notes that Lewis’s argument fails. A change in the shape
of a temporally located object is not a matter of its changing relations to other
changing temporally located things. But it doesn’t follow that this change isn’t a
matter of the object’s changing relations to times, to temporal locations themselves.
This is simply under-determined by our experience of change.
Likewise, a change in the shape or mass of a temporally located object is not a
matter of its changing relations to other temporally located things that don’t consti-
tute it. But it doesn’t follow that this change isn’t a matter of the object’s changing
relations to other temporally located things that do constitute it. This too, is under-
determined by our experience of change. In fact, it seems reasonable to think that a
change in an object’s mass is a certain kind of change in the matter that constitutes
it.
Identity Conditions 232
I know of no other reason to deny that Tibbles’s having a mass is a matter of his
being constituted by something with that mass. We are therefore free to maintain
that it is, and hence to reject premiss (7) of the argument against Relativised In-
heritance. But a sense of unease remains; for there is an important sense in which
Tibbles does not possess any of the properties inherited from his constitutors, viz.
the primary sense in which they possess those properties. Although we’ve found
no reason to reject this view, it doesn’t follow that we should accept it. We may still
hope for better. That is what the next section seeks to provide.
4.4.4.4 Non-relational Relativised Inheritance
This section examines a rejection of premiss (8) of the argument against Relativised
Inheritance. According to such views, this:
Tibbles has φ relative to l.
doesn’t imply this:
Tibbles has φ by bearing a relation to φ and l.
We’ve seen that an object’s having a property (or standing in a relation) needn’t be
analysable into the bearing of a relation between object and property (or between
relata and relation). Let S be a structure with constituents s1, . . . , sn. Then we’ve
seen that S needn’t be analysable into the bearing of a relation amongst s1, . . . , sn.
Not all structure is relational structure. This section argues that the structures
posited by Relativised Inheritance needn’t be relational structures.
Consider the structure that Tibbles, 5kg and l enter into when Tibbles has 5kg
relative to l. We’ll call this structure S. We want to know: is S analysable into
the bearing of a relation between Tibbles, 5kg and l? I can see three unpersuasive
reasons to think that it is.
The first reason to endorse the relational analysis of S begins with the following
necessary equivalence:
Tibbles has 5kg relative to l iff: l has 5kg and l constitutes Tibbles.
Necessary equivalence makes it formally permissible to posit an analysis. But it
doesn’t mandate it. Consider the following necessary equivalences:
Identity Conditions 233
Socrates exists iff {Socrates} exists.
Grass is green or grass is not green iff 2 + 2 = 4.
The r.h.s. of the former certainly doesn’t analyse the left; and the left is an, at
best, controversial analysis of the right. Neither side of the the second is plausibly
any kind of analysis of the other. Necessary equivalence therefore doesn’t imply
analysis.
The second reason appeals to the principle:
Every structure with more than two constituents is analysable into the bear-
ing of a relation amongst those constituents.
Since S has three constituents, this implies that S is analysable into the bearing of
a relation amongst them. Unfortunately for the defender of this second argument,
this principle is false. When a relation holds between two relata, the resulting
structure has three constituents. But we’ve already seen that not all such structures
are analysable into the bearing of further relations amongst their constituents.
The third and final reason appeals to two theses:
(i) S has three constituents.
(ii) An instantiation of a monadic property φ is a structure with (at most) two
constituents: φ itself and the object that has φ.
It follows that S is not an instantiation of a monadic property: either 5kg is really a
relation, or S is analysable into the bearing of a relation amongst its constituents.
This argument fails because (ii) is false: the number of entities involved in an
instantiation is no guide to the degree of the property instantiated. (for discussion
see MacBride, 2005b, §2.) Two examples to make the general point: identity is a
dyadic relation whose instantiations can only involve one other entity; at least as
large as is a dyadic relation some of whose instantiations involve two other enti-
ties, and others of which involve only one. Three examples to make the point for
the special case of monadic properties: arranged in a circle is a monadic (plural
collective) property whose instantiations involve different numbers of entities on
different occasions (furthermore, those instantiations never involve only one other
Identity Conditions 234
entity); being a property and being self-instantiating are properties some of who’s
instantiation feature only one constituent, namely that property itself. Since the
number of objects involved in an instantiation alongside the property or relation
instantiated is no guide to adicity, thesis (ii) above is false.
Adicity is not a feature of how many objects are involved in instantiations, but
of the kinds of resemblance and difference marked by a property or relation. Our
thesis of Relativised Inheritance concerns the number of entities involved in in-
stantiations of inherited properties, not varieties of resemblance and difference.
That thesis is therefore compatible with the monadicity of 5kg and the claim that S
isn’t analysable into the bearing of a relation between Tibbles, 5kg and l.
I can find no other argument for premiss (8) from p.226. We are therefore free
to reject it, and the argument against Relativised Inheritance along with it.
4.4.5 Property-possession: concluding remarks
This section began with three accounts of property-inheritance. We settled on the
relativisation of inherited properties to constituters and saw two defensible ver-
sions of this view. One analyses x’s possession of φ by l into: l possesses φ and
constitutes x. The other rejects that analysis: relativised instantiational structure
is a sui generis variety of structure, not analysable in other terms. Although both
views are defensible, the second carries an advantage over the first: the connection
between Tibbles and his inherited properties is not mediated by any relation. We
turn now to Lewis’s puzzle of constitutional vagueness. One way of developing our
proposal (discussed in §4.5.5) in order to accommodate this puzzle will favour the
first form of Relativised Inheritance over the second.
4.5 Vagueness
Up to now, we’ve focused on Unger’s puzzle of too many candidates and assumed
that it’s clear which the best candidates are. This section relaxes this assumption
and extends our proposal to Lewis’s puzzle of vague constitution and borderline
candidates.
§4.5.1 begins by arguing from an unmodified version of our proposal to unclar-
Identity Conditions 235
ity in constitution, mereology and inherited properties. §4.5.2 then argues that
despite this unclarity, our proposal is incompatible with the Sharpening View of
constitutional vagueness, unless it is modified in some way. We can draw two con-
clusions from this. Firstly, Unger’s puzzle is a source of unclarity in constitution,
but that unclarity is not a form of vagueness: Unger’s puzzle of too many best
candidates is not a puzzle of full-blown vagueness, though it does give rise to a
more limited form of unclarity. Secondly, our proposal must be modified in or-
der to accommodate constitutional vagueness. Different kinds of response to the
initial argument for incompatibility provide different strategies for extending our
proposal to constitutional vagueness. To this end, we close §4.5.2 by uncovering
a hidden assumption within that argument and three potential responses. These
responses are developed in §§4.5.3–4.5.5.
4.5.1 Unclarity in constitution, inheritance and parthood
We’ve proposed that Tibbles is constituted by each of his Many, and that he has his
inherited properties only relative to these constituters. This section argues from
both theses to unclarity in constitution, inherited properties and parthood. The
next section argues that this is not a form of vagueness, but a more limited vari-
ety of unclarity. There is work to be done before we can accommodate genuine
vagueness.
4.5.1.1 Unclarity in constitution
‘Constitution’ was introduced as a name for whichever relation occupies the pre-
theoretic “making up” role. Although this role is implicitly defined by our use of
language as a whole, there is some leeway about just how closely the constitution-
relation fits this role: a reasonably good fit may be good enough. Our thesis of
multiple-constitution brings deviation from this role. Two examples: we may ordi-
narily talk about the wool from which a jumper is made; or we might say, in an only
slightly more theoretical vein, that a statue is in the same place as the particles that
make it up. These definite descriptions manifest the assumption that constitution
is unique, thereby incorporating that assumption into the implicit definition of the
Identity Conditions 236
constitution-role.
A question now arises: how does the extension of ordinary constitutional vo-
cabulary relate to our postulated relation of multiple-constitution? To avoid con-
fusion, we’ll call this relation con and write as if ‘constitutes’ were part of ordinary
English. We’ll also restrict these notions to a single kind of ordinary object, specifi-
cation of which we’ll tend to leave tacit. Then our question is: what is the relation
between con and the extension of ‘constitutes’? The puzzle is that we use ‘consti-
tutes’ as if it were one-one, while our proposal is that con is many-one.
Let us treat relations as sets of ordered pairs, and ignore non-constitutional vo-
cabulary for simplicity. Then we can frame two hypotheses about the relationship
between ‘constitutes’ and con:
Identity An interpretation s is intended iff J‘constitutes’Ks = con.
Inclusion An interpretation s is intended iff:
(i) J‘constitutes’Ks ⊆ con; and
(ii) If 〈x, z〉 ∈ J‘constitutes’Ks and 〈y, z〉 ∈ J‘constitutes’Ks, then x = y; and
(iii) If 〈x, z〉 ∈ con, then, for some y : 〈y, z〉 ∈ J‘constitutes’Ks.
According to Identity, there is exactly one intended interpretation of ‘constitutes’:
the relation con. Since con is many-one, descriptions like ‘the clay that makes up
the statue’ are improper. Sentences featuring them are therefore be untrue.
According to Inclusion, there are many intended interpretations of ‘constitutes’.
Condition (i) ensures that constitutional vocabulary aims at describing the con
facts: if x constitutes y, then con(x, y). Condition (ii) ensures that constitution is
one-one. Condition (iii) ensures that everything that should have a constituter—
everything to which something bears con—does have a constituter. (i)–(iii) to-
gether ensure that the intended interpretations of ‘constitutes’ are the minimal
deviations from con to give a relation that fits the constitution-role. ‘Constitutes’
has many intended interpretations because there are many such minimal devia-
tions. This brings unclarity in constitution without making ‘the matter of Tibbles’
improper.
Identity Conditions 237
Two semantic pressures must be reconciled. The first is our use of constitutional
vocabulary to describe the object-matter relation con. The second is our use of
constitutional vocabulary as if constitution were unique. Our proposal brings these
pressures into conflict by making con many-one. The Identity hypothesis resolves
in favour of the first pressure. This results in improper descriptions and untruth.
The Inclusion hypothesis resolves in favour of the second pressure. This results in
many intended interpretations of ‘constitutes’, and hence unclarity in constitution.
Two arguments suggest that the second kind of resolution wins. The first is a
methodological argument: it makes true a greater proportion of ordinary talk, and
theories that do so are ceteris paribus preferable to theories that don’t. The second is
a metaphysical argument: semantic values are (to a significant extent) determined
by which sentences ordinary speakers hold true. Our solution to Unger’s puzzle
therefore entails unclarity in ordinary constitutional vocabulary.
4.5.1.2 Unclarity in inheritance
This section argues from our proposal to unclarity in ascriptions of inherited prop-
erties. This unclarity will infect mass, location, shape, and any other property that
can vary across Tibbles’s constituters (provided it is inherited by cats). Given the
following connection between property-ascription and predication, this unclarity
will extend beyond explicit property-ascriptions:
x has the property of being F iff x is F.
Ordinary property-ascription isn’t relativised to a constituter. Since our pro-
posal relativises inherited properties to constituters, the following challenge arises:
to convert the relativised ascriptions into truth-conditions for un-relativised as-
criptions.
An un-relativised notion R(x) is most naturally obtained from a relativised
one R∗(x, y) by closing the y-position in R∗. We might use a quantifier or other
variable-binding operator; or we might substitute y for a singular term. We’ll
opt for this second alternative. Each intended interpretation s selects a lump ls
from amongst Tibbles’s constituters and supplies the following truth-condition to
property-ascriptions:
Identity Conditions 238
‘Tibbles has φ’ is s-true iff Tibbles has φ relative to ls.
Since Tibbles has many constituters, many intended interpretations of property-
ascription result. Since Tibbles’s constituters have different masses, the truth-
values of ascriptions of inherited properties to Tibbles vary across intended in-
terpretations. The result is unclarity about Tibbles’s mass. Likewise for other in-
herited properties.
Although other accounts of property-ascription are possible, this one is the best;
for only it can respect the truth of:
Tibbles has the same location and mass as his constituter.
This is secured by the following penumbral connection:
x s-satisfies ‘constitutes Tibbles’ iff: Tibbles has φ is s-true iff Tibbles has φ
relative to x.
This connection ensures that the lump that features in the s-truth-conditions for
property-ascriptions is the lump that s counts as Tibbles’s constituter. Unclarity
in constitution induces unclarity in inherited property-ascription via analytic con-
nections between constitution and inheritance.
4.5.1.3 Unclarity in Parthood
Finally, we can argue from unclarity in constitution to unclarity in the mereology
of ordinary objects.
On one kind of view, object-mereology is definable via constitution and a part-
hood relation on matter:
x is part of an ordinary object o iff x is part of the lump that constitutes o.
Given this, ‘is part of’ has many intended interpretations if ‘constitutes’ does:
unclarity in constitution entails unclarity in object-mereology.
This kind of view is controversial. Although my heart is part of me, it doesn’t
seem to be part of any lump of matter; my heart’s matter is part of my matter, but
my heart itself isn’t. Me and my heart are thus a counterexample to the equivalence
Identity Conditions 239
above. Approaches to object-mereology based around that equivalence are incom-
patible with hierarchical conceptions on which my heart isn’t part of any lump of
matter.21 Still, parthood and constitution are connected:
If x is part of o, then x’s constituter is part of o’s constituter.
An object’s parts must be constituted by parts of its matter. Since different sharpen-
ings count different candidates as Tibbles’s constituter the result is unclarity about
Tibbles’s microscopic parts. Unclarity in constitution entails unclarity in object-
mereology.
4.5.2 Constitutional vagueness?
This section argues that despite the argument from our proposal to unclarity in
constitutional vocabulary, there is work to be done before we can accommodate
constitutional vagueness.
§4.5.2.1 begins by arguing that our proposal is, as it stands, incompatible with
a Sharpening-theoretic account of constitutional vagueness. This shows that the
unclarity argued for in the previous section is a more limited phenomenon than
genuine vagueness. Our proposal must be extended somehow in order to accom-
modate constitutional vagueness. We’ll organise our discussion of such exten-
sions around different kinds of response to the initial argument for incompatibil-
ity. To this end §4.5.2.2 identifies a hidden assumption behind that argument, and
§4.5.2.3 sketches three kinds of response. These are investigated in §§4.5.3–4.5.5.
The first response modifies neither our proposal nor the Sharpening View, but re-
jects the hidden assumption. The second retains the assumption and the Sharp-
ening View without modifying our proposal, but explains away the appearance of
constitutional vagueness. (A notational variant on this second response endorses
an account of constitutional vagueness other than the Sharpening View.) The third
21 We might respond with an alternative account of object-mereology: x is part of o iff the matter of
x is part of the matter of o. If this is acceptable to the defender of the hierarchical view, the argument
from unclarity in constitution to unclarity in object-mereology goes through as before. Difficulties
remain however: the sleeve of a woolen jumper is not part of the thread from which the jumper is
made, though the matter of the sleeve is part of the matter of the thread. The following argument in
the text shows that we can sidestep these issues.
Identity Conditions 240
retains the assumption and the Sharpening View, but modifies our proposal by al-
lowing gradual constitution and hence (what the Sharpening theorist regards as
genuine) constitutional vagueness.
4.5.2.1 Sorites-susceptibility and higher-order borderline cases
This section argues that our proposal requires modification in order to accommo-
date the Sharpening View’s conception of constitutional vagueness. We proceed by
arguing that, as it currently stands, our proposal is incompatible with that concep-
tion. Subsequent sections examine potential modifications of our proposal to avoid
this argument, and hence to accommodate constitutional vagueness.
On our proposal, many lumps bear the relation con to Tibbles. §4.5.1 argued
that each of these lumps a borderline case of a cat-constituter. Consider a Sorites
series S such that (i) S begins with lumps that clearly don’t constitute Tibbles, (ii)
S terminates with lumps that bear con to Tibbles, and (iii) for each element x of
the series, its successor x′ differs only very marginally from x in respects relevant
to bearing con to Tibbles. The following Sorites principle is intuitively plausible:
∀x(¬x constitutes a cat → ¬x′ constitutes a cat)(9)
The Sharpening theorist offers two explanations for why these Sorites principles
are attractive, despite their being provably false (§2.5.2). We now argue that our
proposal is (as it currently stands) incompatible with both these explanations, and
hence incompatible with a Sharpening-theoretic account of the Sorites-susceptibility
of ‘constitutes’.
The Sharpening theorist’s first explanation for the attraction of (9) appeals to
their conception of vagueness as the result of imposing a non-gradual classification
onto a gradual transition. When we do so, nearby points in the transition will differ
very little both in those respects involved in the transition, and in their relations to
our linguistic behaviour. A Sorites principle is a natural, though incorrect, way of
articulating this.
Our proposal can’t accommodate this first explanation because con is non-
gradual. Adjacent candidates in our Sorites series S can (and will) differ signifi-
cantly w.r.t. bearing con to Tibbles. (§4.5.5 modifies our proposal to allow that con
Identity Conditions 241
is gradual.)
The Sharpening theorist’s second explanation for the attraction of (9) begins by
observing that no instantiation of its negation is assertable because each is border-
line. The Sharpening theorist then attributes to typical speakers a mistaken slip
from the unassertability of these instantiations to their falsity, and thereby to the
truth of (9).
On the face of it, this first explanation should be applicable to (9). Instantiations
of its negation are of the form:
¬a constitutes a cat ∧ a′ constitutes a cat
§4.5.1 argued that the best candidates (the ones that bear con to Tibbles) are all
borderline cat-constituters. So each sentence of the above form is either clearly false
or borderline, and hence unassertable. So the Sharpening theorist’s attribution of a
mistaken slip from unassertability to falsity should explain the attraction of (9).
Matters are not quite so clear-cut. The present case is relevantly unlike a typ-
ical Sorites. Although the series S terminates with the best candidates to be cat-
constituters, none of them is a clear cat-constituter (despite it being clear that one
of them is a cat-constituter). Since these are the best cases and each has everything
that could be desired in order to be a case—each bears con to Tibbles—they might
quite easily be mistaken for clear cases (as the reasoning behind Unger’s puzzle
suggests that they are). Indeed, they only fail to be clear cases because of the con-
flicting semantic pressures governing constitutional vocabulary that we described
in §4.5.1. Unlike typical cases of vagueness, they aren’t borderline cases because
our use of language privileges no one classificatory boundary, but because two fea-
tures of that use conflict. If this is right, then when a is the last non-candidate,
there’s a sense in which ‘a′ constitutes a cat’ is clearly true; the sense in which our
use of ‘constitutes’ aims at con. Since ‘¬a constitutes a cat’ is then also clearly true,
the result is a clearly true instance of the form displayed above. The Sharpening
View’s explanation for the attraction of Sorites principles therefore doesn’t seem to
extend to principle (9).
These arguments show that there’s work to be done before our proposal can
accommodate a Sharpening-theoretic account of the Sorites-susceptibility of ‘con-
Identity Conditions 242
stitutes’. There’s also work to be done before we can accommodate a Sharpening-
theoretic account of higher-order vagueness in ‘constitutes’.
§2.9.9 developed an account of higher-order vagueness that appeals to metase-
mantic gradualness: a series of interpretations, each of which fits the meaning-
determining facts only slightly less well than its predecessor, gives rise to many
intended interpretations of ‘intended interpretation’, and many intended interpre-
tations of ‘intended interpretation of ‘intended interpretation’ ’, and so on. But
consider a sharpening s on which the last non-candidate in the series satisfies ‘con-
stitutes Tibbles’: the s-extension of ‘constitutes’ isn’t a sub-relation of con, in vi-
olation of our first condition on intended interpretations of ‘constitutes’ (p.236).
Other things being equal, s therefore fits the meaning-determining facts signifi-
cantly less well than interpretations on which the extension of ‘constitutes’ is a
sub-relation of con. This limit on metasemantic gradualness prevents any second-
order borderline cases from separating the clear non-cases from the first-order bor-
derline cases. Our proposal therefore needs modifying before it can accommodate
the Sharpening theorist’s account of higher-order constitutional vagueness.
§§4.5.3–4.5.5 generalise our proposal to accommodate constitutional vague-
ness. They do so by examining three kinds of response to these arguments. The
next section begins by identifying a hidden assumption on which these arguments
rely.
4.5.2.2 A hidden assumption
This section identifies a hidden assumption in the previous section’s argument for
the incompatibility of our proposal with a Sharpening-theoretic account of consti-
tutional vagueness. The assumption is a conception of content-determination akin
to that of Lewis (1983a, 1984).
According to Lewis, content is determined by (at least) two features of inter-
pretations: how well they fit our linguistic behaviour, and the Eligibility of the
semantic values they assign to our vocabulary. Eligibility is a measure of intrinsic
suitability to be meant. Typically, and certainly in Lewis’s view, the Eligibility-
ordering is identified with the naturalness-ordering.22
22 To accommodate singular terms, the naturalness-ordering on properties needs extending to an
Identity Conditions 243
Consider the problem facing the Sharpening theorist’s first account of the Sorites
principle (9). The problem was that since con is non-gradual, some adjacent candi-
dates in the Sorites series S will differ significantly in respects relevant to whether
they satisfy ‘constitutes Tibbles’: one but not the other will bear con to Tibbles.
Why should this matter to how well suited those candidates are to satisfy ‘con-
stitutes Tibbles’? Those candidates won’t differ significantly in their relations to
our use of constitutional vocabulary. The only alternative answer seems to be that
interpretations that make the extension of ‘constitutes’ a sub-relation of con are
ceteris paribus more Eligible than those that don’t.
Consider the problem facing the Sharpening theorist’s second account of the
Sorites. The problem was that since the distinction between those candidates that
bear con to Tibbles and those that don’t isn’t vague, and hence a significant sense
in which there are no borderline candidates. This provides a significant sense in
which some sentence of the following form is clearly true:
¬a constitutes Tibbles ∧ a′ constitutes Tibbles.
Unless interpretations on which the s-satisfier of ‘constitutes Tibbles’ bears con to
Tibbles fit the meaning-determining facts significantly better, ceteris paribus, than
those on which it doesn’t, this argument fails. It’s unclear what could justify that,
other than appeal to the Eligibility of con.
Finally, consider the problem for the Sharpening theorist’s account of higher-
order vagueness. This assumed that the following suffices, ceteris paribus, for a sig-
nificant difference w.r.t. how well interpretations s, t fit the meaning-determining
facts: the s-extension of ‘constitutes’ is a sub-relation of con, though the t-extension
of ‘constitutes’ isn’t. What justifies this, if not the Lewisian conception of content-
determination? Such differences seem insignificant w.r.t. fit with use. Appeal to a
significant difference w.r.t. Eligibility seems to be the only alternative.
ordering on objects. Lewis (1983a, p.49) suggests that we do so by appeal to how well their bound-
aries are demarcated by natural properties.
Identity Conditions 244
4.5.2.3 Three kinds of response
We’ve seen that the Lewisian conception of content-determination is assumed by
the argument in §4.5.2.1 for the incompatibility of our proposal with the Sharp-
ening View’s account of constitutional vagueness. This section distinguishes three
kinds of resistance to that argument. Each provides one way of extending our pro-
posal to accommodate constitutional vagueness.23
(i) Reject the Lewisian conception of content-determination.
(ii) Deny that constitution is vague, and explain away the appearance that it is.
(iii) Develop an account of gradual constitution.
Response (i) retains our proposal and the Sharpening View without modification;
the arguments to show that some modification is needed are rejected instead. Re-
sponse (ii) accepts the arguments for the incompatibility of our proposal with con-
stitutional vagueness, taking them to show that constitutional vagueness is im-
possible. Advocates of this response must explain away the appearance of con-
stitutional vagueness in a way that doesn’t extend to all other cases of (apparent)
vagueness, and thereby undermine the Sharpening View. Response (iii) also ac-
cepts that the problems are genuine, but takes them to show instead that our so-
lution to Unger’s puzzle will not do as it stands; the goal is to make con more like
the properties and relations relevant to typical cases of vagueness. The following
sections consider these in turn. We won’t come to a settled view about which is
preferable; each is defensible, though each has its costs.
4.5.3 Content-determination without Eligibility
This section presents an account of constitutional vagueness that rejects Lewis’s
Eligibility-based conception of content-determination, leaving the Sharpening View
and our response to Lewis’s puzzle unmodified. We begin with some concerns
about the Lewisian account of content-determination.23 A fourth option isn’t considered here: reject the Sharpening View of vagueness wholesale in
favour of an alternative.
Identity Conditions 245
4.5.3.1 Against Eligibility
Although popular, the Lewisian conception of content-determination is somewhat
mysterious. Its best, and probably only, motivation is to respond to Kripke and
Putnam’s arguments for scepticism about meaning (Kripke, 1982; Putnam, 1980).
If there are other, better solutions, then the view is unmotivated. We can’t investi-
gate the alternatives here, but it’s worth noting this way in which Lewis’s view is a
hostage to theoretical fortune.
The view comes in two varieties, depending on whether Eligibility is identified
with naturalness or not. We’ll raise some worries about both varieties.
Consider the view that identifies Eligibility with naturalness. We should ask:
why does this connection hold? Why is a more natural property a better candi-
date semantic value than a less natural one, other things being equal? As Lewis
(1983a, pp.54–5) makes clear, the answer isn’t that we intend to use language to
mark reasonably natural distinctions. Not only is it highly dubious that we have
such intentions, but that answer presupposes an account of content-determination
for intentions; yet the arguments for meaning-scepticism apply to thought and
language both. No alternative account of the Eligibility-naturalness connection
is forthcoming. The result is a surprising and unexplained connection between a
property’s naturalness and its suitability to be expressed by a predicate. This con-
nection comes not from an investigation into the nature of meaning, but a desire to
block a problematic argument. Maybe we should hope for no more than this, but
it is hard to see the result as a unified theoretical package.
Consider now the view that distinguishes Eligibility from naturalness. So what
is Eligibility? There seems no alternative independently motivated ordering on
candidate semantic values whose identification with Eligibility would be any less
mysterious than that of naturalness. So the Eligibility-ordering must be taken as a
sui generis kind of semantic fact: some potential meanings are just better meanings
than others. On the one hand, this doesn’t address Kripke and Putnam’s sceptical
challenges, so much as simply insist that there is a response to them. On the other
hand, it blocks an account of semantic facts in broadly naturalistic or physicalistic
terms.
Identity Conditions 246
We haven’t shown that Lewis’s Eligibility-based account of content-determination
is false. We have shown however, that its motivation is tenuous and it either (i)
brings mysterious connections between seemingly disparate kinds of fact, or (ii)
blocks a naturalistic account of semantics. These are good reasons to be scepti-
cal about it. And once that scepticism is in place, we should also be sceptical of
the arguments purporting to show that our proposal needs modifying in order to
accommodate a Sharpening-theoretic account of constitutional vagueness.
4.5.3.2 Constitutional vagueness without Eligibility
Rejecting Lewis’s account of content-determination undermines an argument to
show that our proposal needs modifying before it can accommodate constitutional
vagueness. It doesn’t follow that our proposal can accommodate that vagueness.
This section sketches an account.
Suppose that hair h is clearly part of Tibbles at time t1, and has fallen out by
time t2. Let T be the lump that constitutes Tibbles at t1; let T−h be T excluding
the matter of h. By t2, T−h constitutes Tibbles and T is a scattered object. We’ll
assume for simplicity that Unger’s puzzle doesn’t arise at t1 or at t2: only T bears
con to Tibbles at t1, and only T−h bears con to Tibbles at t2. We’ll also assume that
h is Tibbles’s only borderline part at any time between t1 and t2, and that Tibbles
undergoes no changes other than those consequent on his loss of h.
When h is a perfectly balanced borderline part of Tibbles, both T and T−h are
equally good (and good enough) candidates to constitute Tibbles; they both then
bear con to Tibbles. Lewis’s puzzle of borderline constituters thus induces Unger’s
puzzle of too many best candidates. We want to expand on this to accommodate
the Sorites and higher-order vagueness. Our strategy is to mirror the Sharpening
theorist’s account of typical non-constitutional vagueness.
As h falls out, it gradually becomes less causally integrated with the rest of
Tibbles. Underlying this gradually weakening causal connection, is non-gradual
variation in con: it holds from T to Tibbles at t1, from both T and T−h to Tibbles
at some intermediate time(s) tn, and only from T−h to Tibbles at t2. The problem
was that this imposes sharp boundaries on ‘constitutes’. But if we reject the role of
Identity Conditions 247
Eligibility in content-determination, then this non-gradual/sharp variation in con
needn’t translate into sharp boundaries in ‘constitutes’. Our use of ‘constitutes’ is
sensitive to the gradually varying causal and spatial relations between h and T−h,
not the non-gradual variation in con. Small variations in these respects bring only
small variations in fit with our use of ‘constitutes’: our use of ‘constitutes’ imposes
a non-gradual classification onto this gradual series without privileging any one
point in the series over all others. Hence, from the Sharpening theorist’s perspec-
tive, just the same features that lead to vagueness in ‘red’, ‘old’ and ‘tall’ lead to
vagueness in ‘constitutes’. Without a role for Eligibility, the cases are alike, and
there’s no bar to applying the Sharpening View of vagueness. We can therefore
accommodate vague constitution without modifying our proposal or the Sharpen-
ing View, and without an abundance of cats, provided we reject Lewis’s account of
content-determination.
4.5.4 Limiting constitutional unclarity
We’ve got one account of constitutional vagueness in place that modifies neither
our proposal nor the Sharpening View. Since that view turns on rejecting Lewis’s
Eligibility-based account of content-determination, it won’t be acceptable to all. So
this section develops a different way of extending our proposal to constitutional
vagueness.
The view developed here accepts the arguments for the incompatibility of our
solution to Unger’s puzzle with the Sharpening View of constitutional vagueness.
This is taken to show that constitution cannot be vague.24 The task is to explain
away the appearance of constitutional vagueness. This is the goal of §4.5.4.1. We’ll
do so by adopting an epistemicist strategy. Two further challenges then arise:
If there cannot be borderline cases to the borderline cases, why allow border-
line cases of constitution at all? Why not have sharp boundaries at the first
level if we’re going to have them anywhere (especially somewhere so close as
the second level)?
24 A notational variant draws the alternative conclusion that vagueness is not a uniform phe-
nomenon.
Identity Conditions 248
Why doesn’t this account generalise to all vagueness, and thereby undermine
the Sharpening View?
These are addressed in §§4.5.4.2–4.5.4.3.
4.5.4.1 First challenge: the Sorites
Consider a Sorites series on the constitution of Tibbles by T, originating at (i) a
time t1 when h was clearly part of Tibbles, who was then constituted by T, and
terminating with (ii) a time tn when h clearly wasn’t part of Tibbles, who was then
constituted by T−h. The first challenge is to explain why the following is intuitively
plausible, although provably false, and to do so despite the non-gradualness of con:
(10) ∀ti(T constitutes Tibbles at ti → T constitutes Tibbles at ti+1)
We can meet this challenge by co-opting an epistemicist strategy. Our exposition of
this strategy will ignore the unclarity in constitution that arises when both T and
T−h bear con to Tibbles; in other words, we ’ll assume that con holds first from T
to Tibbles, and then from T−h to Tibbles, and never from both to Tibbles. It follows
that Unger’s puzzle never arises. Nothing of substance turns on this, but it simpli-
fies exposition and makes our task harder by providing a clear counterexample to
(10). We’ll also assume that everything is named.
Our strategy is as follows. First, we’ll explain why no instance of the following
is knowable (to beings like ourselves), despite one of them being clearly true:
(11) T constitutes Tibbles at tα ∧ ¬T constitutes Tibbles at tα+1
Then we’ll postulate a (mistaken) slip from the unknowability of these instances
to their falsity, and provide an explanation of why we make this mistake. Now, if
every instance of (11) is false, then so is:
(12) ∃ti(T constitutes Tibbles at ti ∧ ¬T constitutes Tibbles at ti+1)
And if that is false, then (10) is true. The result is an (invalid but natural) argument
from the unknowability of instances of (11) to the truth of the Sorites principle
(10). We’ll explain the attraction of that Sorites principle by attributing this kind
of reasoning to ordinary speakers. Let’s turn to the details.
Identity Conditions 249
Why is no instance of (11) knowable (to beings like ourselves), despite one be-
ing clearly true? Suppose we know all the facts about causal integration, spatial
separation and the like that concern T and h. Suppose also that we know exactly
what kinds of change Tibbles survives (under a one-level mode of presentation).
These are all the facts relevant to con. So unless the extension of con is knowable
on the basis of these facts, it isn’t knowable at all (to beings like ourselves). But
the extension of con isn’t knowable on that basis unless we know how the basis
bears on con. Since we don’t know that, and its not clear how we might find it out,
any means of inferring the extension of con from these facts would be no better
than a guess, even if it gave accurate results; and even an accurate guess doesn’t
yield knowledge. So we can’t know the extension of con, and hence can’t know any
instance of (11).
The next task is to explain the slip from the unknowability of any instance of
(11) to the falsity of each. Note first that we either do or could in principle know all
the facts relevant to the truth of instances of (11). Yet no amount of investigation
into those facts would reveal which instance was true. Since this unknowability
isn’t the result of our own limitations, there must be no truth there to know. So
each instance of (11) is false; so (12) is false; so the Sorites principle (10) is true.
Attributing this kind of reasoning to ordinary speakers allows us to explain the
slip from the unknowability of each instance of (11) to the truth of (10), despite
its falsity. The flaw in the argument is that there’s a kind of fact relevant to the
extension of con that we don’t know: how what we do know bears on con. This ap-
proach to the Sorites thus attributes forgetfulness or ignorance about the existence
of these facts to ordinary speakers.25 By doing so, we can explain the attraction of
constitutional Sorites principles without appeal to (what the Sharpening theorist
regards as genuine) vagueness in constitution, and without modifying our response
to Unger. The following two sections elaborate this view by responding to the two
challenges immediately preceding this section.
25 One candidate explanation for this ignorance might attribute a form of microphysicalism to
ordinary speakers: all relations between macroscopic and microscopic entities are revealed by mi-
crophysical descriptions.
Identity Conditions 250
4.5.4.2 Second challenge: borderline precision
§4.5.1 argued from our thesis of multiple-constitution to borderline cases of consti-
tution: the many best candidates that all bear con to Tibbles are borderline cases
of cat-constituters. The present view posits a sharp boundary between the clear
non-cat-constituters and the borderline cat-constituters. Why is this preferable to
a sharp boundary between the constituters and the non-constituters? The answer
is that it isn’t.
On the present view, there is a significant sense in which the constituter/non-
constituter distinction is non-vague: there are no borderline cases of con. Our
argument from multiple-constitution (i.e. from con being many-one) to borderline
constitution didn’t appeal to vagueness or the imposition of an absolute classifica-
tion onto a gradual transition. It appealed instead to the reconciliation of conflict-
ing semantic pressures when determining the extension of ‘constitutes’: we speak
as if constitution were one-one; con isn’t one-one; yet con is the best candidate to
occupy the constitution-role. If it weren’t for these conflicting pressures, we could
dispense with unclarity in constitution: the lumps con-related to Tibbles would
clearly constitute him, and everything else would clearly fail to.
The moral is that the present approach to (apparent) constitutional vagueness
doesn’t treat first- and second-order vagueness differently. It denies the existence
of both, and hence posits borderline cases in response to neither. It does treat
first- and second-order unclarity differently, but that’s because the argument for the
former doesn’t extend to an argument for the latter: the unclarity in constitution
that arises from Unger’s puzzle is not a form of vagueness.
4.5.4.3 Third challenge: generalisation to other cases
The Sharpening theorist who endorses this approach must explain why it doesn’t
extend to all other forms of vagueness: why adopt the Sharpening View at all, if
an alternative is adequate? This challenge can be met by pointing to a disanalogy
with typical cases of vagueness, like ‘red’.
Beneath h’s gradual working loose lies a non-gradual and highly Eligible dis-
tinction between the times when T bears con to Tibbles, and those when it doesn’t.
Identity Conditions 251
con serves as a “reference magnet” for ordinary constitutional vocabulary’, impos-
ing precision on ‘constitutes’ despite our messy use of language. In this respect, the
Sharpening theorist should claim, ‘constitutes’ is unlike ‘red’, ‘tall’ and ‘young’.
Gradual variation in shade, height and age do not mask any highly Eligible non-
gradual distinction; nothing plays the role of con in imposing a sharp boundary on
our messy use of ‘red’, ‘tall’ and ‘young’. If this is correct, then the present strategy
of explaining away the appearance of constitutional vagueness doesn’t extend to
those cases.
4.5.5 Gradual constitution
We’ve seen that we can accommodate constitutional vagueness by rejecting Lewis’s
Eligibility-based account of content-determination (§4.5.3). We’ve also seen that
we can explain away the appearance of constitutional vagueness if we retain that
account of content-determination (§4.5.4). This section presents our third and final
method for accommodating constitutional vagueness. It generalises our solution to
Unger’s puzzle by allowing con to be gradual. Vagueness in ‘constitutes’ can then
be treated in just the same way as for any other form of vagueness: the result of
our imposing an absolute classification onto a gradual transition. §4.5.5.1 intro-
duces the proposal. §4.5.5.2 turns to an objection. §4.5.5.3 closes by examining the
proposal’s interaction with relativised instantiation.
4.5.5.1 The proposal
According to the Sharpening theorist, vagueness results from our imposition of
absolute classifications onto a gradual world. The problem with accommodating
constitutional vagueness within our solution to Unger’s puzzle was that con is non-
gradual. Were con gradual, there would be no problem. So why not let con be
gradual, and thereby eliminate the problem?
What exactly would it be for con to be gradual? Consider the gradual transition
of shades from orange to red on a colour chart. This gradualness consists in two
things. One is the instantiation of many different determinates of the determinable
Identity Conditions 252
colour. The other is an ordering on these determinates.26 We can apply this to con,
as follows.
There are many determinate con-relations, each belonging to the same deter-
minable. There is also an ordering on these con-determinates. For simplicity we’ll
assume that this ordering is total and dense. Nothing of substance turns on this,
but it allows us to write as if a single con-relation held to a degree d, where d is a
real number in the interval [0, 1]; larger numbers represent stronger constitutional-
connections (greater elements in the ordering on con-determinates).
Our two-level proposal may seem to face a problem with gradual constitution;
for that view treats constitution as a function from matter to ordinary objects, and
functional-application is non-gradual. However, for each function f there is a func-
tional relation R f such that:
f (x) = y iff R f (x, y).
An ontology of functions is thus eliminable in favour of functional relations and
function signs governed by the rule:
p f (α)q denotes the unique object y such that R f (x, y), where x is the referent
of α.
Then we can rewrite two-level criteria thus:
∀x∀y∃z∃z′[(R f (x, z) ∧ R f (y, z′) ∧ z = z′)↔ R(x, y)]
In our two-level proposal, R f is con (as restricted to an ordinary kind K). That
proposal can therefore allow gradual constitution.
Once con is gradual, the Sharpening View can be applied. Neither our use
‘constitutes’ nor con itself privileges some unique degree of con over all others.
This gives rise to many intended interpretations of ‘constitutes’. Sorites principles
are attractive because they seem to report the absence of relevant differences be-
tween successive cases in a Sorites series. Higher-order vagueness arises because
the metasemantic facts that determine the intended interpretations of ‘intended
interpretation of ‘constitutes’ ’ are gradual.26 On typical colour charts, the left-right ordering of exemplars of shades matches the ordering on
shades themselves. The gradual transition amongst shades is mirrored in their layout on the chart.
The ordering on shades is also multi-dimensional, but we’ll ignore this complication for simplicity.
Identity Conditions 253
4.5.5.2 A limit on higher-order borderline cases?
Although gradualness in con may allow vagueness in ‘constitutes’, this proposal
may appear to impose a limit on the extent of higher-order vagueness. The prob-
lem is that the distinction between standing in con to no degree and doing so to
some degree is non-gradual. This may seem to limit metasemantic gradualness,
and hence also higher-order vagueness.
To illustrate the problem, let R, R∗ be candidate extensions for ‘constitutes’ that
differ only as follows:
Let x be a lump that bears con to Tibbles to degree 0. Let y be a lump that
bears con to Tibbles to some degree only just greater than 0. R and R∗ both
hold from y to Tibbles, but only R∗ holds from x to Tibbles.
The question is: does this suffice to make R∗ significantly less Eligible than R? If so,
then interpretations that assign R∗ to ‘constitutes’ will fit the meaning-determining
facts significantly less well than those that assign R to ‘constitutes’. This lim-
its metasemantic gradualness: no series of interpretations, each of which fits the
meaning-determining facts only slightly less well than its predecessor, connects
interpretations of the following kinds:
The s-extension of ‘constitutes Tibbles’ includes something that bears con to
Tibbles to degree 0.
The s-extension of ‘constitutes Tibbles’ includes something that bears con to
Tibbles to some degree only just greater than 0.
The result is that objects that don’t bear con to Tibbles at all will be absolutely
clearly non-constituters of Tibbles, and no borderline cases will separate them from
everything else.
There are two kinds of response we might take. The first denies that this limit
on higher-order vagueness is an objectionable limit. The second denies that the
argument for this limit is sound. We take them in turn.
Is this limit on higher-order vagueness objectionable? One reason to think not
appeals to a version of the view in §4.5.4: this limit isn’t the result of our use
Identity Conditions 254
of language, but is imposed by the underlying facts about con. In particular, it’s
imposed by the non-gradual distinction between standing in con to some degree,
and standing in con to no degree. In order to be objectionable, a limit on higher-
order vagueness would have to result from our use of language. Since this one
doesn’t, it isn’t objectionable.
The second kind of response denies that the difference between R and R∗ suf-
fices for a significant difference in their Eligibility to be interpretations of ‘consti-
tutes’. It’s hard to argue either way, given how little is known about Eligibility.
It’s even more difficult if the Eligibility-ordering is identified with the naturalness-
ordering. One reason is that the naturalness-ordering is defined using perfect nat-
uralness, which is supposed to be primitive. Another reason is that it’s unclear
how perfect naturalness determines the naturalness-ordering.27 But still, the dif-
ference between being con-related to Tibbles to some arbitrarily small degree and
not being con-related to Tibbles at all doesn’t look like a very significant objective
difference. Consider the change from one state to the other. No gradual shift may
accompany this change, but it doesn’t follow that it’s a very significant change: it
may not correspond to any major variation in the intrinsic nature of the object in
question. It needn’t even be a greater change than a change in the degree to which
something bears con to Tibbles, if the ordering on con-determinates isn’t dense.
In light of these considerations, both the following are doubtful: (a) allowing
con to be gradual limits the extent of higher-order vagueness; (b) any limits on
higher-order vagueness resulting from our gradual account of con are objection-
able limits.
4.5.5.3 Relativised property-possession
§4.4 defended the following view: if Tibbles inherits a property φ from a consti-
tuter l, then he doesn’t have φ simpliciter, but only relative to l. How does this
interact with gradual constitution? There seem to be two suggestions:
Tibbles has φ relative to l iff l bears con to Tibbles to some degree greater
27 Lewis (1986b, p.61) suggests the following: φ is more natural than ψ iff φ can be reached by a
shorter and (or?) less-complicated chain of definability from the perfect naturals than can ψ. But
what is the objective standard for complexity and length of a definition?
Identity Conditions 255
than n and has φ.
Tibbles has φ relative to l to degree d iff l both bears con to Tibbles to degree
d and has φ.
We should reject the first. On that view, if l bears con to Tibbles to less than degree
n, then there’s no sense in which Tibbles has the same mass, shape, location and so
on as l. It’s mysterious how l could then count as even remotely constitutionally
connected to Tibbles. So let’s consider the second view.
On our preferred account of relativised possession (§4.4.4.4), property, lump
and object all enter into a single structure; this structure isn’t analysable into a
relation’s obtaining amongst its constituents. When combined with gradual consti-
tution, this yields as many different varieties of these structures as there are con-
determinates. There seem to be two problems with this. The first is that it brings a
massive increase in our theory’s primitive ideology. The second is that it’s unclear
what these structures all have in common: why do they all count as relativised-
possession-structures? Were they analysable using an instantiation relation I, then
we could appeal to different determinates of the determinable I. But that analysis
is just what our preferred view of relativised possession denies. The defender of
gradual constitution should therefore prefer the alternative account of relativised
possession (§4.4.4.3). On that view, relativised possession is analysable in terms
of constitution and the properties of matter. This allows us to take the r.h.s. of
the second biconditional above as an analysis of the left, and hence of relativised
possession to a degree without any ideological cost. The defender of gradual con-
stitution can then accommodate vague constitution, higher-order borderline cases
of constitution, and degrees of relativised possession simply by appeal to the grad-
ualness of con. This completes our third account of vague constitution. The Sharp-
ening theorist who is prepared to allow gradual constitution can accommodate the
vagueness of constitution without an abundance of cats.
Identity Conditions 256
4.6 Conclusion
This chapter presented several solutions to the Problem of the Many. Each devel-
ops the thesis that change is explanatorily prior to constitution. §4.2 presented
one-level and two-level identity criteria as ways of developing this view. The key
difference between these two views lies in what kinds of change they claim take
priority over constitution: changes in the persisting object itself, or changes in its
matter. Although the two views aren’t in direct competition, §4.3.5 argued that
the one-level view is preferable, without ruling out the two-level view entirely.
The choice between these views also doesn’t affect our solution to Unger’s puzzle:
Tibbles is constituted by each of the best candidates on his mat. Both the one-
level view and the two-level allow us to mount direct arguments for this claim.
This shows that, unlike Lowe and Johnston’s proposals, ours is not merely an arbi-
trary collection of theses designed to invalidate the arguments for many cats. §4.4
finished the exposition of our solution to Unger’s puzzle by relativising Tibbles’s
inherited properties, like mass, colour and location, to the matter from which he
inherits them.
§4.5 closed with a discussion of unclarity and vagueness. We argued from our
proposal to unclarity in constitution, mereology and inherited properties. These
arguments exploit a mismatch between linguistic structure and the structure of
the underlying facts in order to locate many equally suitable interpretations of the
vocabulary in question. We then argued that this unclarity isn’t genuine vagueness,
but another form of linguistic unclarity: Unger’s puzzle is not directly a puzzle of
vagueness. We closed with three ways of extending our solution to Unger’s puzzle
to constitutional vagueness. The first relied on a rejection of Lewis’s Eligibility-
based conception of content-determination. The second rejected constitutional
vagueness and attempted to explain away its appearance without undermining the
Sharpening View. The third modified our solution to Unger’s puzzle by allowing
constitution to be a gradual matter. Each of these views has its own costs and ben-
efits, which there isn’t space here to evaluate properly. Whichever of these views
we prefer, the result is a unified solution to both Unger’s and Lewis’s Problems of
the Many on which there is only ever one cat on Tibbles’s mat, and a conception of
Identity Conditions 257
ordinary objects from which this solution emerges naturally.
258
Bibliography
Barnett, David (2008). “Indeterminacy and incomplete definitions”. The Journal of
Philosophy, 55: 167–91.
Blackburn, Simon and Simmons, Keith, eds. (1999). Truth. OUP.
Boolos, George (1984). “To be is to be the value of a variable (or some values of
some variables)”. The Journal of Philosophy, 81: 430–49.
Chisholm, Roderick M. (1976). Person and Object: A Metaphysical Study. George
Allen & Unwin.
Cook, Roy T. (2002). “Vagueness and mathematical precision”. Mind, 111: 225–
247.
Dever, Josh (2009). “The disunity of truth”. In Robert J. Stainton and Christopher
viger, eds., Compositionality, Context and Semantic Values: Essays in Honour of
Ernie Lepore, chapter 7. Springer Netherlands.
Dorr, Cian (2003). “Vagueness without ignorance”. Philosophical Perspectives 17:
Language and Philosophical Linguistics, 83–113.
Edgington, Dorothy (1997). “Vagueness by degrees”. In Rosanna Keefe and Peter
Smith, eds., Vagueness: a reader, chapter 16. MIT Press.
—— (2010). “Possible knowledge of unknown truth”. Synthese, 173: 41–52.
Eklund, Matti (2008). “Deconstructing ontological vagueness”. Canadian Journal
of Philosophy, 38: 117–40.
Bibliography 259
Etchemendy, John (1990). The Concept of Logical Consequence. Cambridge Mass.,
London England: Harvard University Press.
Evans, Gareth (1978). “Can there be vague objects?” Analysis, 38: 208.
Field, Hartry (1974). “Quine and the correspondence theory”. The Philosophical
Review, 83: 200–28.
—— (2008). Saving Truth From Paradox. Oxford: OUP.
Fine, Kit (1975). “Vagueness, truth and logic”. Synthese 30, 265–300.
—— (1999). “Things and their parts”. Midwest Studies in Philosophy, 23: 61–74.
—— (2000). “A counter-example to Locke’s Thesis”. The Monist, 83: 357–61.
—— (2003). “The non-identity of a material thing and its matter”. Mind, 112:
195–234.
—— (2008). “Coincidence and form”. The Aristotelian Society: Supplementary Vol-
ume, 82: 101–18.
Geach, Peter (1980). Reference and Generality. 3rd. edition. Ithaca, NY: Cornell
University Press.
Hawley, Katherine (2001). How Things Persist. Oxford, New York: OUP.
Hawthorne, John (2006a). “Epistemicism and semantic plasticity”. Oxford Studies
in Metaphysics, 2: 289–324. Reprinted in Hawthorne (2006b, ch.9).
—— (2006b). Metaphysical Essays. Clarendon Press, Oxford: OUP.
—— (2006c). “Plenitude, convention, and ontology”. In Hawthorne (2006b, ch.3).
—— (2006d). “Three-dimensionalism”. In Hawthorne (2006b, ch.5).
Hazen, Allen (1979). “Counterpart-theoretic semantics for modal logic”. Journal of
Philosophy, 76: 319–38.
Hirsch, Eli (1982). The Concept of Identity. OUP.
Bibliography 260
Horsten, Leon (2010). “Impredicative Identity Criteria”. Philosophy and Phe-
nomenological Research, 80: 411–39.
Hudson, Hud (2001). A Materialist Metaphysics of the Human Person. Ithaca, NY:
Cornell University Press.
Hughes, Christopher (1986). “Is a thing just the sum of its parts?” Proceedings of
the Aristotelian Society, 86: 213–33. Reprinted in Noonan (1993, pp.79–99).
Hyde, Dominic (1997). “From heaps and gaps to heaps of gluts”. Mind, 108: 641–
80.
Johnston, Mark (1992). “Constitution is not identity”. Mind, 101: 89–105.
Reprinted in Rea (1997, pp.44–62).
Jones, Nicholas K. (forthcoming). “Williams on supervaluationism and logical re-
visionism”. Journal of Philosophy.
Keefe, Rosanna (2000). Theories of Vagueness. Cambridge: Cambridge University
Press.
Keefe, Rosanna and Smith, Peter, eds. (1997). Vagueness: A Reader. MIT Press.
Kripke, Saul A. (1982). Wittgenstein on Rules and Private Language. Oxford: Black-
wells.
LePore, Ernest and Fodor, Jerry (1996). “What cannot be evaluated cannot be eval-
uated, and it cannot be supervalued either”. Journal of Philosophy, 93: 513–36.
Lewis, David (1968). “Counterpart theory and quantified modal logic”. The Journal
of Philosophy, 65: 113–26. Reprinted in Lewis (1983b, ch.3).
—— (1969). Convention: A Philosophical Study. Cambridge, Mass.: Harvard Uni-
versity Press.
—— (1970a). “General Semantics”. Synthese, 22: 18–67.
—— (1970b). “How to define theoretical terms”. The Journal of Philosophy, 67:
427–46. Reprinted in Lewis (1983b, ch.6).
Bibliography 261
—— (1971). “Counterparts of persons and their bodies”. The Journal of Philosophy,
68: 203–11. Reprinted in Lewis (1983b, ch.4).
—— (1976). “Survival and identity”. In Amelie O. Rorty, ed., The Identities of
Persons, 17–40. University of California Press. Reprinted in Lewis (1983b, ch.5).
—— (1983a). “New work for a theory of universals”. The Australasian Journal of
Philosophy, 61: 343–77. Reprinted in Lewis (1999, pp.8–55).
—— (1983b). Philosophical Papers: Volume 1. New York, Oxford: OUP.
—— (1984). “Putnam’s Paradox”. The Australasian Journal of Philosophy, 62: 221–
36. Reprinted in Lewis (1999, pp.56–77).
—— (1986a). Counterfactuals. revised edition. Cambridge MA: Harvard University
Press.
—— (1986b). On the Plurality of Worlds. Malden MA, Carlton Victoria, Oxford:
Blackwell Publishing.
—— (1988a). “Rearrangement of particles: reply to Lowe”. Analysis, 48: 65–72.
Reprinted in Lewis (1999, ch.11).
—— (1988b). “Vague identity: Evans misunderstood”. Analysis, 48: 128–30.
Reprinted in Keefe and Smith (1997, ch.18).
—— (1991). Parts of Classes. Oxford: Basil Blackwell.
—— (1993a). “Many, but almost one”. In Ontology, Causality and Mind: Essays
on the Philosophy of D.M. Armstrong, 23–37. Cambridge: Cambridge University
Press. Reprinted in Lewis (1999, ch.9).
—— (1993b). “Mathematics is megethology”. Philosophia Mathematica, 1: 3–23.
—— (1999). Papers in Metaphysics and Epistemology. Cambridge University Press.
—— (2002). “Tensing the copula”. Mind, 111: 1–13.
—— (2009). “Ramseyan humility”. In David Braddon-Mitchell & Robert Nola, ed.,
Conceptual Analysis and Philosophical Naturalism, chapter 9. MIT Press.
Bibliography 262
Loux, Michael J. and Zimmerman, Dean (2003). Oxford Handbook of Metaphysics.
Oxford, New York: OUP.
Lowe, E. J. (1982a). “On being a cat”. Analysis, 42: 174–77.
—— (1982b). “Reply to Geach”. Analysis, 42: 31.
—— (1982c). “The paradox of 1,001 cats”. Analysis, 42: 27–30.
—— (1989). “What is a criterion of identiy?” The Philosophical Quarterly, 39: 1–21.
—— (1991). “One-level versus two-level identity criteria”. Analysis, 51: 192–4.
—— (1995). “The problem of the many and the vagueness of constitution”. Analy-
sis, 55: 179–82.
MacBride, Fraser (2001). “Four new ways to change your shape”. Australasian
Journal of Philosophy, 79: 81–9.
—— (2005a). “Review of Herbert Hochberg, Introducing Analytic Philosophy: Its
Sense and its Nonsense 1879–2002”. Dialectica, 59: 81–87.
—— (2005b). “The particular-universal distinction: a dogma of metaphysics?”
Mind, 114: 565–614.
Markosian, Ned (1998). “Brutal composition”. Philsophical Studies, 92: 211–249.
McFetridge, Ian (1990). “Logical necessity: some issues”. In Logical Necessity and
Other Essays, number 10 in Aristotelian Society Series, 135–54. London: Aris-
totelian Society. Edited by John Haldane and Roger Scruton.
McGee, Vann and McLaughlin, Brian (1994). “Distinctions without a difference”.
The Southern Journal of Philosophy, 33: 203–261.
—— (1998). “Review of Williamson (1994)”. Linguistics and Philosophy, 21: 221–
35.
—— (2000). “The lessons of the Many”. Philosophical Topics, 28: 129–51.
—— (2004). “Logical commitment and semantic indeterminacy: a reply to
Williamson”. Linguistics and Philosophy, 27: 122–36.
Bibliography 263
McKinnon, Neil (2002). “Supervaluations and the problem of the many”. The
Philosophical Quarterly, 52: 320–39.
Merricks, Trenton (2003). Objects and Persons. Clarendon Press: OUP.
Moore, Jospeh (1999). “Propositions without identity”. Nous, 33: 1–29.
Noonan, Harold, ed. (1993). Identity. Dartmouth Publishing Company.
Putnam, Hilary (1980). “Models and reality”. Journal of Symbolic Logic, 45: 464–82.
Quine, W. V. O. (1976). “Worlds away”. The Journal os Philosophy, 73: 859–65.
—— (1981a). Theories and Things. Harvard University Press.
—— (1981b). “What price bivalence?” Journal of Philosophy. Reprinted with
additions in Quine (1981a, ch.3).
Rea, Michael C., ed. (1997). Material Constitution: A Reader. Lanham, Boulder, New
York, Oxford: Rowman & Lanham Publishers inc.
Rumfitt, Ian (1993). “Content and context: the paratactic theory revisited and
revised”. Mind, 102: 429–54.
—— (2002). “Concepts and counting”. Proceedings of the Aristotelian Society, 102:
41–68.
Sainsbury, R. M. (1990). Concepts without boundaries. King’s College London De-
patment of Philosophy. Reprinted in Keefe and Smith (1997, pp.251–64).
Schiffer, Stephen (1998). “Two issues of vagueness”. The Monist, 81: pp.193–214.
—— (2000). “Replies”. Philosophical Issues, 10: 321–43.
Shapiro, Stewart (1998). “Logical consequence: models and modality”. In Matthias
Schirn, ed., The Philosophy of Mathematics Today, 131–56. Clarendon Press: OUP.
—— (2006). Vagueness in Context. Clarendon Press: OUP.
Sharvy, Richard (1980). “A more general theory of definite descriptions”. The
Philosophical Review, 89: 607–24.
Bibliography 264
Sider, Theodore (2001a). Four-Dimensionalism: An Ontology of Persistence and Time.
Oxford, New York: OUP.
—— (2001b). “Maximality and intrinsic properties”. Philosophy and Phenomeno-
logical Research, 63: 357–64.
—— (2003). “Maximality and microphysical supervenience”. Philosophy and Phe-
nomenological Research, 66: 139–49.
Simons, Peter (1987). Parts: A Study in Ontology. Clarendon Press: OUP.
—— (2000). “Continuants and occurrents”. Proceedings of the Aristotelian Society:
Supplementary Volume, 74: 59–75.
—— (2008). “The thread of persistence”. In Christian Kanzian, ed., Persistence,
165–84. Ontos Verlag.
Sorensen, Roy (2000). “Direct reference and vague identity”. Philosophical Topics,
28: 175–94.
—— (2010). “Borderline hermaphrodites: higher-order vagueness by example”.
Mind, 119: 393–408.
Stalnaker, Robert C. (1986). “Counterparts and Identity”. Midwest Studies in Phi-
losophy: Studies in Essentialism, 11: 121–40. Reprinted in Stalnaker (2003, ch.6).
—— (1994). “The interaction of quantification with modality and identity”. In
N. Asher W. Sinnott Armostrong, D. Raffman, ed., Modality, Morality and Belief:
Essays in honor of Ruth Barcan Marcus. Cambridge: Cambridge University Press.
Reprinted in Stalnaker (2003, ch.8).
—— (2003). Ways a World Might Be: Metaphysical and Anti-Metaphysical Essays.
Clarendon Press: OUP.
Tarski, Alfred (1936). “Über den begriff der logischen folgerung”. Actes du Congrès
International de Philosophie Scientifique, 7: 1–11. Translated as “On the concept
of logical consequence” in Tarski (1956).
—— (1956). Logic, Semantic, Metamathematics. Oxford: Clarendon Press.
Bibliography 265
Tye, Michael (1996). “Fuzzy realism and the problem of the many”. Philosophical
Studies, 81: 215–25.
Unger, Peter K. (1979). “Why there are no people”. Midwest Studies in Philosophy,
IV: 177–222. Reprinted in Unger (2006b, pp.53-111).
—— (1980). “The problem of the many”. Midwest Studies in Philosophy, 5: 411–67.
Reprinted in Unger (2006b, pp.113–83).
—— (2006a). All the Power in the World. Oxford, New York: OUP.
—— (2006b). Philosophical Papers: Volume 2. Oxford, New York: OUP.
van Inwagen, Peter (1990). Material Beings. Ithaca, London: Cornell University
Press.
Weatherson, Brian (2003). “Many many problems”. The Philosophical Quarterly, 53:
481–501.
—— (2009). “The Problem of the Many”. In Edward N. Zalta, ed., The Stanford
Encyclopedia of Philosophy, summer 2009 edition. Accessed October 2009.
URL http://plato.stanford.edu/archives/sum2009/entries/
problem-of-many/
Wiggins, David (1968). “On being in the same place at the same time”. The Philo-
sophical Review, 77: 90–95. Reprinted in Rea (1997, ch.1).
—— (1980). Sameness and Substance. Oxford: Basil Blackwell.
—— (2001). Sameness and Substance Renewed. Cambridge Universit Press.
Williams, J. R. G. (2006). “An argument for the many”. Proceedings of the Aris-
totelian Society, 106: 411–19.
—— (2008). “Supervaluationism and logical revisionism”. Journal of Philosophy,
55: 192–212.
Williamson, Timothy (1990). Identity and Discrimination. Basil Blackwell.
—— (1994). Vagueness. London, New York: Routledge.
Bibliography 266
—— (1997). “Imagination, stipulation and vagueness”. Philosophical Issues, 8: 215–
28.
—— (1999). “On the structure of higher-order vagueness”. Mind, 108: 127–43.
—— (2003). “Vagueness in Reality”. In Loux & Zimmerman Loux and Zimmerman
(2003), 690–715.
—— (2004). “Reply to McGee and McLaughlin”. Linguistics and Philosophy, 27:
113–22.
—— (2007). The Philosophy of Philosophy. Blackwell Publishing.
Wright, Crispin (1976). “Language-mastery and the sorites paradox”. In G Evans
and J McDowell, eds., Truth and Meaning: Essays in Semantics, 223–47. Oxford,
Clarendon Press: OUP. Reprinted in Keefe and Smith (1997, ch.10).
—— (1999). “Truth: a traditional debate reviewed”. Canadian Journal of Philosophy,
24. Reprinted in Blackburn and Simmons (1999, ch.13).
Zimmerman, Dean W. (1995). “Theories of Masses and Problems of Constitution”.
The Philosophical Review, 104: 53–110.